2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/kmod.h>
111 #include <linux/module.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130 #include <trace/events/napi.h>
131 #include <linux/pci.h>
133 #include "net-sysfs.h"
135 /* Instead of increasing this, you should create a hash table. */
136 #define MAX_GRO_SKBS 8
138 /* This should be increased if a protocol with a bigger head is added. */
139 #define GRO_MAX_HEAD (MAX_HEADER + 128)
142 * The list of packet types we will receive (as opposed to discard)
143 * and the routines to invoke.
145 * Why 16. Because with 16 the only overlap we get on a hash of the
146 * low nibble of the protocol value is RARP/SNAP/X.25.
148 * NOTE: That is no longer true with the addition of VLAN tags. Not
149 * sure which should go first, but I bet it won't make much
150 * difference if we are running VLANs. The good news is that
151 * this protocol won't be in the list unless compiled in, so
152 * the average user (w/out VLANs) will not be adversely affected.
169 #define PTYPE_HASH_SIZE (16)
170 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
172 static DEFINE_SPINLOCK(ptype_lock
);
173 static struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
174 static struct list_head ptype_all __read_mostly
; /* Taps */
177 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
180 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
182 * Writers must hold the rtnl semaphore while they loop through the
183 * dev_base_head list, and hold dev_base_lock for writing when they do the
184 * actual updates. This allows pure readers to access the list even
185 * while a writer is preparing to update it.
187 * To put it another way, dev_base_lock is held for writing only to
188 * protect against pure readers; the rtnl semaphore provides the
189 * protection against other writers.
191 * See, for example usages, register_netdevice() and
192 * unregister_netdevice(), which must be called with the rtnl
195 DEFINE_RWLOCK(dev_base_lock
);
196 EXPORT_SYMBOL(dev_base_lock
);
198 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
200 unsigned hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
201 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
204 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
206 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
209 static inline void rps_lock(struct softnet_data
*sd
)
212 spin_lock(&sd
->input_pkt_queue
.lock
);
216 static inline void rps_unlock(struct softnet_data
*sd
)
219 spin_unlock(&sd
->input_pkt_queue
.lock
);
223 /* Device list insertion */
224 static int list_netdevice(struct net_device
*dev
)
226 struct net
*net
= dev_net(dev
);
230 write_lock_bh(&dev_base_lock
);
231 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
232 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
233 hlist_add_head_rcu(&dev
->index_hlist
,
234 dev_index_hash(net
, dev
->ifindex
));
235 write_unlock_bh(&dev_base_lock
);
239 /* Device list removal
240 * caller must respect a RCU grace period before freeing/reusing dev
242 static void unlist_netdevice(struct net_device
*dev
)
246 /* Unlink dev from the device chain */
247 write_lock_bh(&dev_base_lock
);
248 list_del_rcu(&dev
->dev_list
);
249 hlist_del_rcu(&dev
->name_hlist
);
250 hlist_del_rcu(&dev
->index_hlist
);
251 write_unlock_bh(&dev_base_lock
);
258 static RAW_NOTIFIER_HEAD(netdev_chain
);
261 * Device drivers call our routines to queue packets here. We empty the
262 * queue in the local softnet handler.
265 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
266 EXPORT_PER_CPU_SYMBOL(softnet_data
);
268 #ifdef CONFIG_LOCKDEP
270 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
271 * according to dev->type
273 static const unsigned short netdev_lock_type
[] =
274 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
275 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
276 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
277 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
278 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
279 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
280 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
281 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
282 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
283 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
284 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
285 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
286 ARPHRD_FCFABRIC
, ARPHRD_IEEE802_TR
, ARPHRD_IEEE80211
,
287 ARPHRD_IEEE80211_PRISM
, ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
,
288 ARPHRD_PHONET_PIPE
, ARPHRD_IEEE802154
,
289 ARPHRD_VOID
, ARPHRD_NONE
};
291 static const char *const netdev_lock_name
[] =
292 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
293 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
294 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
295 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
296 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
297 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
298 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
299 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
300 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
301 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
302 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
303 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
304 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
305 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
306 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
307 "_xmit_VOID", "_xmit_NONE"};
309 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
310 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
312 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
316 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
317 if (netdev_lock_type
[i
] == dev_type
)
319 /* the last key is used by default */
320 return ARRAY_SIZE(netdev_lock_type
) - 1;
323 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
324 unsigned short dev_type
)
328 i
= netdev_lock_pos(dev_type
);
329 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
330 netdev_lock_name
[i
]);
333 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
337 i
= netdev_lock_pos(dev
->type
);
338 lockdep_set_class_and_name(&dev
->addr_list_lock
,
339 &netdev_addr_lock_key
[i
],
340 netdev_lock_name
[i
]);
343 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
344 unsigned short dev_type
)
347 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
352 /*******************************************************************************
354 Protocol management and registration routines
356 *******************************************************************************/
359 * Add a protocol ID to the list. Now that the input handler is
360 * smarter we can dispense with all the messy stuff that used to be
363 * BEWARE!!! Protocol handlers, mangling input packets,
364 * MUST BE last in hash buckets and checking protocol handlers
365 * MUST start from promiscuous ptype_all chain in net_bh.
366 * It is true now, do not change it.
367 * Explanation follows: if protocol handler, mangling packet, will
368 * be the first on list, it is not able to sense, that packet
369 * is cloned and should be copied-on-write, so that it will
370 * change it and subsequent readers will get broken packet.
374 static inline struct list_head
*ptype_head(const struct packet_type
*pt
)
376 if (pt
->type
== htons(ETH_P_ALL
))
379 return &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
383 * dev_add_pack - add packet handler
384 * @pt: packet type declaration
386 * Add a protocol handler to the networking stack. The passed &packet_type
387 * is linked into kernel lists and may not be freed until it has been
388 * removed from the kernel lists.
390 * This call does not sleep therefore it can not
391 * guarantee all CPU's that are in middle of receiving packets
392 * will see the new packet type (until the next received packet).
395 void dev_add_pack(struct packet_type
*pt
)
397 struct list_head
*head
= ptype_head(pt
);
399 spin_lock(&ptype_lock
);
400 list_add_rcu(&pt
->list
, head
);
401 spin_unlock(&ptype_lock
);
403 EXPORT_SYMBOL(dev_add_pack
);
406 * __dev_remove_pack - remove packet handler
407 * @pt: packet type declaration
409 * Remove a protocol handler that was previously added to the kernel
410 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
411 * from the kernel lists and can be freed or reused once this function
414 * The packet type might still be in use by receivers
415 * and must not be freed until after all the CPU's have gone
416 * through a quiescent state.
418 void __dev_remove_pack(struct packet_type
*pt
)
420 struct list_head
*head
= ptype_head(pt
);
421 struct packet_type
*pt1
;
423 spin_lock(&ptype_lock
);
425 list_for_each_entry(pt1
, head
, list
) {
427 list_del_rcu(&pt
->list
);
432 printk(KERN_WARNING
"dev_remove_pack: %p not found.\n", pt
);
434 spin_unlock(&ptype_lock
);
436 EXPORT_SYMBOL(__dev_remove_pack
);
439 * dev_remove_pack - remove packet handler
440 * @pt: packet type declaration
442 * Remove a protocol handler that was previously added to the kernel
443 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
444 * from the kernel lists and can be freed or reused once this function
447 * This call sleeps to guarantee that no CPU is looking at the packet
450 void dev_remove_pack(struct packet_type
*pt
)
452 __dev_remove_pack(pt
);
456 EXPORT_SYMBOL(dev_remove_pack
);
458 /******************************************************************************
460 Device Boot-time Settings Routines
462 *******************************************************************************/
464 /* Boot time configuration table */
465 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
468 * netdev_boot_setup_add - add new setup entry
469 * @name: name of the device
470 * @map: configured settings for the device
472 * Adds new setup entry to the dev_boot_setup list. The function
473 * returns 0 on error and 1 on success. This is a generic routine to
476 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
478 struct netdev_boot_setup
*s
;
482 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
483 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
484 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
485 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
486 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
491 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
495 * netdev_boot_setup_check - check boot time settings
496 * @dev: the netdevice
498 * Check boot time settings for the device.
499 * The found settings are set for the device to be used
500 * later in the device probing.
501 * Returns 0 if no settings found, 1 if they are.
503 int netdev_boot_setup_check(struct net_device
*dev
)
505 struct netdev_boot_setup
*s
= dev_boot_setup
;
508 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
509 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
510 !strcmp(dev
->name
, s
[i
].name
)) {
511 dev
->irq
= s
[i
].map
.irq
;
512 dev
->base_addr
= s
[i
].map
.base_addr
;
513 dev
->mem_start
= s
[i
].map
.mem_start
;
514 dev
->mem_end
= s
[i
].map
.mem_end
;
520 EXPORT_SYMBOL(netdev_boot_setup_check
);
524 * netdev_boot_base - get address from boot time settings
525 * @prefix: prefix for network device
526 * @unit: id for network device
528 * Check boot time settings for the base address of device.
529 * The found settings are set for the device to be used
530 * later in the device probing.
531 * Returns 0 if no settings found.
533 unsigned long netdev_boot_base(const char *prefix
, int unit
)
535 const struct netdev_boot_setup
*s
= dev_boot_setup
;
539 sprintf(name
, "%s%d", prefix
, unit
);
542 * If device already registered then return base of 1
543 * to indicate not to probe for this interface
545 if (__dev_get_by_name(&init_net
, name
))
548 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
549 if (!strcmp(name
, s
[i
].name
))
550 return s
[i
].map
.base_addr
;
555 * Saves at boot time configured settings for any netdevice.
557 int __init
netdev_boot_setup(char *str
)
562 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
567 memset(&map
, 0, sizeof(map
));
571 map
.base_addr
= ints
[2];
573 map
.mem_start
= ints
[3];
575 map
.mem_end
= ints
[4];
577 /* Add new entry to the list */
578 return netdev_boot_setup_add(str
, &map
);
581 __setup("netdev=", netdev_boot_setup
);
583 /*******************************************************************************
585 Device Interface Subroutines
587 *******************************************************************************/
590 * __dev_get_by_name - find a device by its name
591 * @net: the applicable net namespace
592 * @name: name to find
594 * Find an interface by name. Must be called under RTNL semaphore
595 * or @dev_base_lock. If the name is found a pointer to the device
596 * is returned. If the name is not found then %NULL is returned. The
597 * reference counters are not incremented so the caller must be
598 * careful with locks.
601 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
603 struct hlist_node
*p
;
604 struct net_device
*dev
;
605 struct hlist_head
*head
= dev_name_hash(net
, name
);
607 hlist_for_each_entry(dev
, p
, head
, name_hlist
)
608 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
613 EXPORT_SYMBOL(__dev_get_by_name
);
616 * dev_get_by_name_rcu - find a device by its name
617 * @net: the applicable net namespace
618 * @name: name to find
620 * Find an interface by name.
621 * If the name is found a pointer to the device is returned.
622 * If the name is not found then %NULL is returned.
623 * The reference counters are not incremented so the caller must be
624 * careful with locks. The caller must hold RCU lock.
627 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
629 struct hlist_node
*p
;
630 struct net_device
*dev
;
631 struct hlist_head
*head
= dev_name_hash(net
, name
);
633 hlist_for_each_entry_rcu(dev
, p
, head
, name_hlist
)
634 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
639 EXPORT_SYMBOL(dev_get_by_name_rcu
);
642 * dev_get_by_name - find a device by its name
643 * @net: the applicable net namespace
644 * @name: name to find
646 * Find an interface by name. This can be called from any
647 * context and does its own locking. The returned handle has
648 * the usage count incremented and the caller must use dev_put() to
649 * release it when it is no longer needed. %NULL is returned if no
650 * matching device is found.
653 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
655 struct net_device
*dev
;
658 dev
= dev_get_by_name_rcu(net
, name
);
664 EXPORT_SYMBOL(dev_get_by_name
);
667 * __dev_get_by_index - find a device by its ifindex
668 * @net: the applicable net namespace
669 * @ifindex: index of device
671 * Search for an interface by index. Returns %NULL if the device
672 * is not found or a pointer to the device. The device has not
673 * had its reference counter increased so the caller must be careful
674 * about locking. The caller must hold either the RTNL semaphore
678 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
680 struct hlist_node
*p
;
681 struct net_device
*dev
;
682 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
684 hlist_for_each_entry(dev
, p
, head
, index_hlist
)
685 if (dev
->ifindex
== ifindex
)
690 EXPORT_SYMBOL(__dev_get_by_index
);
693 * dev_get_by_index_rcu - find a device by its ifindex
694 * @net: the applicable net namespace
695 * @ifindex: index of device
697 * Search for an interface by index. Returns %NULL if the device
698 * is not found or a pointer to the device. The device has not
699 * had its reference counter increased so the caller must be careful
700 * about locking. The caller must hold RCU lock.
703 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
705 struct hlist_node
*p
;
706 struct net_device
*dev
;
707 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
709 hlist_for_each_entry_rcu(dev
, p
, head
, index_hlist
)
710 if (dev
->ifindex
== ifindex
)
715 EXPORT_SYMBOL(dev_get_by_index_rcu
);
719 * dev_get_by_index - find a device by its ifindex
720 * @net: the applicable net namespace
721 * @ifindex: index of device
723 * Search for an interface by index. Returns NULL if the device
724 * is not found or a pointer to the device. The device returned has
725 * had a reference added and the pointer is safe until the user calls
726 * dev_put to indicate they have finished with it.
729 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
731 struct net_device
*dev
;
734 dev
= dev_get_by_index_rcu(net
, ifindex
);
740 EXPORT_SYMBOL(dev_get_by_index
);
743 * dev_getbyhwaddr - find a device by its hardware address
744 * @net: the applicable net namespace
745 * @type: media type of device
746 * @ha: hardware address
748 * Search for an interface by MAC address. Returns NULL if the device
749 * is not found or a pointer to the device. The caller must hold the
750 * rtnl semaphore. The returned device has not had its ref count increased
751 * and the caller must therefore be careful about locking
754 * If the API was consistent this would be __dev_get_by_hwaddr
757 struct net_device
*dev_getbyhwaddr(struct net
*net
, unsigned short type
, char *ha
)
759 struct net_device
*dev
;
763 for_each_netdev(net
, dev
)
764 if (dev
->type
== type
&&
765 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
770 EXPORT_SYMBOL(dev_getbyhwaddr
);
772 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
774 struct net_device
*dev
;
777 for_each_netdev(net
, dev
)
778 if (dev
->type
== type
)
783 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
785 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
787 struct net_device
*dev
, *ret
= NULL
;
790 for_each_netdev_rcu(net
, dev
)
791 if (dev
->type
== type
) {
799 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
802 * dev_get_by_flags_rcu - find any device with given flags
803 * @net: the applicable net namespace
804 * @if_flags: IFF_* values
805 * @mask: bitmask of bits in if_flags to check
807 * Search for any interface with the given flags. Returns NULL if a device
808 * is not found or a pointer to the device. Must be called inside
809 * rcu_read_lock(), and result refcount is unchanged.
812 struct net_device
*dev_get_by_flags_rcu(struct net
*net
, unsigned short if_flags
,
815 struct net_device
*dev
, *ret
;
818 for_each_netdev_rcu(net
, dev
) {
819 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
826 EXPORT_SYMBOL(dev_get_by_flags_rcu
);
829 * dev_valid_name - check if name is okay for network device
832 * Network device names need to be valid file names to
833 * to allow sysfs to work. We also disallow any kind of
836 int dev_valid_name(const char *name
)
840 if (strlen(name
) >= IFNAMSIZ
)
842 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
846 if (*name
== '/' || isspace(*name
))
852 EXPORT_SYMBOL(dev_valid_name
);
855 * __dev_alloc_name - allocate a name for a device
856 * @net: network namespace to allocate the device name in
857 * @name: name format string
858 * @buf: scratch buffer and result name string
860 * Passed a format string - eg "lt%d" it will try and find a suitable
861 * id. It scans list of devices to build up a free map, then chooses
862 * the first empty slot. The caller must hold the dev_base or rtnl lock
863 * while allocating the name and adding the device in order to avoid
865 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
866 * Returns the number of the unit assigned or a negative errno code.
869 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
873 const int max_netdevices
= 8*PAGE_SIZE
;
874 unsigned long *inuse
;
875 struct net_device
*d
;
877 p
= strnchr(name
, IFNAMSIZ
-1, '%');
880 * Verify the string as this thing may have come from
881 * the user. There must be either one "%d" and no other "%"
884 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
887 /* Use one page as a bit array of possible slots */
888 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
892 for_each_netdev(net
, d
) {
893 if (!sscanf(d
->name
, name
, &i
))
895 if (i
< 0 || i
>= max_netdevices
)
898 /* avoid cases where sscanf is not exact inverse of printf */
899 snprintf(buf
, IFNAMSIZ
, name
, i
);
900 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
904 i
= find_first_zero_bit(inuse
, max_netdevices
);
905 free_page((unsigned long) inuse
);
909 snprintf(buf
, IFNAMSIZ
, name
, i
);
910 if (!__dev_get_by_name(net
, buf
))
913 /* It is possible to run out of possible slots
914 * when the name is long and there isn't enough space left
915 * for the digits, or if all bits are used.
921 * dev_alloc_name - allocate a name for a device
923 * @name: name format string
925 * Passed a format string - eg "lt%d" it will try and find a suitable
926 * id. It scans list of devices to build up a free map, then chooses
927 * the first empty slot. The caller must hold the dev_base or rtnl lock
928 * while allocating the name and adding the device in order to avoid
930 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
931 * Returns the number of the unit assigned or a negative errno code.
934 int dev_alloc_name(struct net_device
*dev
, const char *name
)
940 BUG_ON(!dev_net(dev
));
942 ret
= __dev_alloc_name(net
, name
, buf
);
944 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
947 EXPORT_SYMBOL(dev_alloc_name
);
949 static int dev_get_valid_name(struct net_device
*dev
, const char *name
, bool fmt
)
953 BUG_ON(!dev_net(dev
));
956 if (!dev_valid_name(name
))
959 if (fmt
&& strchr(name
, '%'))
960 return dev_alloc_name(dev
, name
);
961 else if (__dev_get_by_name(net
, name
))
963 else if (dev
->name
!= name
)
964 strlcpy(dev
->name
, name
, IFNAMSIZ
);
970 * dev_change_name - change name of a device
972 * @newname: name (or format string) must be at least IFNAMSIZ
974 * Change name of a device, can pass format strings "eth%d".
977 int dev_change_name(struct net_device
*dev
, const char *newname
)
979 char oldname
[IFNAMSIZ
];
985 BUG_ON(!dev_net(dev
));
988 if (dev
->flags
& IFF_UP
)
991 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0)
994 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
996 err
= dev_get_valid_name(dev
, newname
, 1);
1001 ret
= device_rename(&dev
->dev
, dev
->name
);
1003 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1007 write_lock_bh(&dev_base_lock
);
1008 hlist_del(&dev
->name_hlist
);
1009 write_unlock_bh(&dev_base_lock
);
1013 write_lock_bh(&dev_base_lock
);
1014 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1015 write_unlock_bh(&dev_base_lock
);
1017 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1018 ret
= notifier_to_errno(ret
);
1021 /* err >= 0 after dev_alloc_name() or stores the first errno */
1024 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1028 "%s: name change rollback failed: %d.\n",
1037 * dev_set_alias - change ifalias of a device
1039 * @alias: name up to IFALIASZ
1040 * @len: limit of bytes to copy from info
1042 * Set ifalias for a device,
1044 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1048 if (len
>= IFALIASZ
)
1053 kfree(dev
->ifalias
);
1054 dev
->ifalias
= NULL
;
1059 dev
->ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1063 strlcpy(dev
->ifalias
, alias
, len
+1);
1069 * netdev_features_change - device changes features
1070 * @dev: device to cause notification
1072 * Called to indicate a device has changed features.
1074 void netdev_features_change(struct net_device
*dev
)
1076 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1078 EXPORT_SYMBOL(netdev_features_change
);
1081 * netdev_state_change - device changes state
1082 * @dev: device to cause notification
1084 * Called to indicate a device has changed state. This function calls
1085 * the notifier chains for netdev_chain and sends a NEWLINK message
1086 * to the routing socket.
1088 void netdev_state_change(struct net_device
*dev
)
1090 if (dev
->flags
& IFF_UP
) {
1091 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1092 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1095 EXPORT_SYMBOL(netdev_state_change
);
1097 int netdev_bonding_change(struct net_device
*dev
, unsigned long event
)
1099 return call_netdevice_notifiers(event
, dev
);
1101 EXPORT_SYMBOL(netdev_bonding_change
);
1104 * dev_load - load a network module
1105 * @net: the applicable net namespace
1106 * @name: name of interface
1108 * If a network interface is not present and the process has suitable
1109 * privileges this function loads the module. If module loading is not
1110 * available in this kernel then it becomes a nop.
1113 void dev_load(struct net
*net
, const char *name
)
1115 struct net_device
*dev
;
1118 dev
= dev_get_by_name_rcu(net
, name
);
1121 if (!dev
&& capable(CAP_NET_ADMIN
))
1122 request_module("%s", name
);
1124 EXPORT_SYMBOL(dev_load
);
1126 static int __dev_open(struct net_device
*dev
)
1128 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1134 * Is it even present?
1136 if (!netif_device_present(dev
))
1139 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1140 ret
= notifier_to_errno(ret
);
1145 * Call device private open method
1147 set_bit(__LINK_STATE_START
, &dev
->state
);
1149 if (ops
->ndo_validate_addr
)
1150 ret
= ops
->ndo_validate_addr(dev
);
1152 if (!ret
&& ops
->ndo_open
)
1153 ret
= ops
->ndo_open(dev
);
1156 * If it went open OK then:
1160 clear_bit(__LINK_STATE_START
, &dev
->state
);
1165 dev
->flags
|= IFF_UP
;
1170 net_dmaengine_get();
1173 * Initialize multicasting status
1175 dev_set_rx_mode(dev
);
1178 * Wakeup transmit queue engine
1187 * dev_open - prepare an interface for use.
1188 * @dev: device to open
1190 * Takes a device from down to up state. The device's private open
1191 * function is invoked and then the multicast lists are loaded. Finally
1192 * the device is moved into the up state and a %NETDEV_UP message is
1193 * sent to the netdev notifier chain.
1195 * Calling this function on an active interface is a nop. On a failure
1196 * a negative errno code is returned.
1198 int dev_open(struct net_device
*dev
)
1205 if (dev
->flags
& IFF_UP
)
1211 ret
= __dev_open(dev
);
1216 * ... and announce new interface.
1218 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1219 call_netdevice_notifiers(NETDEV_UP
, dev
);
1223 EXPORT_SYMBOL(dev_open
);
1225 static int __dev_close(struct net_device
*dev
)
1227 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1233 * Tell people we are going down, so that they can
1234 * prepare to death, when device is still operating.
1236 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1238 clear_bit(__LINK_STATE_START
, &dev
->state
);
1240 /* Synchronize to scheduled poll. We cannot touch poll list,
1241 * it can be even on different cpu. So just clear netif_running().
1243 * dev->stop() will invoke napi_disable() on all of it's
1244 * napi_struct instances on this device.
1246 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1248 dev_deactivate(dev
);
1251 * Call the device specific close. This cannot fail.
1252 * Only if device is UP
1254 * We allow it to be called even after a DETACH hot-plug
1261 * Device is now down.
1264 dev
->flags
&= ~IFF_UP
;
1269 net_dmaengine_put();
1275 * dev_close - shutdown an interface.
1276 * @dev: device to shutdown
1278 * This function moves an active device into down state. A
1279 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1280 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1283 int dev_close(struct net_device
*dev
)
1285 if (!(dev
->flags
& IFF_UP
))
1291 * Tell people we are down
1293 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1294 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1298 EXPORT_SYMBOL(dev_close
);
1302 * dev_disable_lro - disable Large Receive Offload on a device
1305 * Disable Large Receive Offload (LRO) on a net device. Must be
1306 * called under RTNL. This is needed if received packets may be
1307 * forwarded to another interface.
1309 void dev_disable_lro(struct net_device
*dev
)
1311 if (dev
->ethtool_ops
&& dev
->ethtool_ops
->get_flags
&&
1312 dev
->ethtool_ops
->set_flags
) {
1313 u32 flags
= dev
->ethtool_ops
->get_flags(dev
);
1314 if (flags
& ETH_FLAG_LRO
) {
1315 flags
&= ~ETH_FLAG_LRO
;
1316 dev
->ethtool_ops
->set_flags(dev
, flags
);
1319 WARN_ON(dev
->features
& NETIF_F_LRO
);
1321 EXPORT_SYMBOL(dev_disable_lro
);
1324 static int dev_boot_phase
= 1;
1327 * Device change register/unregister. These are not inline or static
1328 * as we export them to the world.
1332 * register_netdevice_notifier - register a network notifier block
1335 * Register a notifier to be called when network device events occur.
1336 * The notifier passed is linked into the kernel structures and must
1337 * not be reused until it has been unregistered. A negative errno code
1338 * is returned on a failure.
1340 * When registered all registration and up events are replayed
1341 * to the new notifier to allow device to have a race free
1342 * view of the network device list.
1345 int register_netdevice_notifier(struct notifier_block
*nb
)
1347 struct net_device
*dev
;
1348 struct net_device
*last
;
1353 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1359 for_each_netdev(net
, dev
) {
1360 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1361 err
= notifier_to_errno(err
);
1365 if (!(dev
->flags
& IFF_UP
))
1368 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1379 for_each_netdev(net
, dev
) {
1383 if (dev
->flags
& IFF_UP
) {
1384 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1385 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1387 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1388 nb
->notifier_call(nb
, NETDEV_UNREGISTER_BATCH
, dev
);
1392 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1395 EXPORT_SYMBOL(register_netdevice_notifier
);
1398 * unregister_netdevice_notifier - unregister a network notifier block
1401 * Unregister a notifier previously registered by
1402 * register_netdevice_notifier(). The notifier is unlinked into the
1403 * kernel structures and may then be reused. A negative errno code
1404 * is returned on a failure.
1407 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1412 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1416 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1419 * call_netdevice_notifiers - call all network notifier blocks
1420 * @val: value passed unmodified to notifier function
1421 * @dev: net_device pointer passed unmodified to notifier function
1423 * Call all network notifier blocks. Parameters and return value
1424 * are as for raw_notifier_call_chain().
1427 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1430 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1433 /* When > 0 there are consumers of rx skb time stamps */
1434 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1436 void net_enable_timestamp(void)
1438 atomic_inc(&netstamp_needed
);
1440 EXPORT_SYMBOL(net_enable_timestamp
);
1442 void net_disable_timestamp(void)
1444 atomic_dec(&netstamp_needed
);
1446 EXPORT_SYMBOL(net_disable_timestamp
);
1448 static inline void net_timestamp_set(struct sk_buff
*skb
)
1450 if (atomic_read(&netstamp_needed
))
1451 __net_timestamp(skb
);
1453 skb
->tstamp
.tv64
= 0;
1456 static inline void net_timestamp_check(struct sk_buff
*skb
)
1458 if (!skb
->tstamp
.tv64
&& atomic_read(&netstamp_needed
))
1459 __net_timestamp(skb
);
1463 * dev_forward_skb - loopback an skb to another netif
1465 * @dev: destination network device
1466 * @skb: buffer to forward
1469 * NET_RX_SUCCESS (no congestion)
1470 * NET_RX_DROP (packet was dropped, but freed)
1472 * dev_forward_skb can be used for injecting an skb from the
1473 * start_xmit function of one device into the receive queue
1474 * of another device.
1476 * The receiving device may be in another namespace, so
1477 * we have to clear all information in the skb that could
1478 * impact namespace isolation.
1480 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1485 if (!(dev
->flags
& IFF_UP
) ||
1486 (skb
->len
> (dev
->mtu
+ dev
->hard_header_len
))) {
1490 skb_set_dev(skb
, dev
);
1491 skb
->tstamp
.tv64
= 0;
1492 skb
->pkt_type
= PACKET_HOST
;
1493 skb
->protocol
= eth_type_trans(skb
, dev
);
1494 return netif_rx(skb
);
1496 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1499 * Support routine. Sends outgoing frames to any network
1500 * taps currently in use.
1503 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1505 struct packet_type
*ptype
;
1507 #ifdef CONFIG_NET_CLS_ACT
1508 if (!(skb
->tstamp
.tv64
&& (G_TC_FROM(skb
->tc_verd
) & AT_INGRESS
)))
1509 net_timestamp_set(skb
);
1511 net_timestamp_set(skb
);
1515 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1516 /* Never send packets back to the socket
1517 * they originated from - MvS (miquels@drinkel.ow.org)
1519 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1520 (ptype
->af_packet_priv
== NULL
||
1521 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1522 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
1526 /* skb->nh should be correctly
1527 set by sender, so that the second statement is
1528 just protection against buggy protocols.
1530 skb_reset_mac_header(skb2
);
1532 if (skb_network_header(skb2
) < skb2
->data
||
1533 skb2
->network_header
> skb2
->tail
) {
1534 if (net_ratelimit())
1535 printk(KERN_CRIT
"protocol %04x is "
1537 ntohs(skb2
->protocol
),
1539 skb_reset_network_header(skb2
);
1542 skb2
->transport_header
= skb2
->network_header
;
1543 skb2
->pkt_type
= PACKET_OUTGOING
;
1544 ptype
->func(skb2
, skb
->dev
, ptype
, skb
->dev
);
1551 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1552 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1554 void netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
1556 unsigned int real_num
= dev
->real_num_tx_queues
;
1558 if (unlikely(txq
> dev
->num_tx_queues
))
1560 else if (txq
> real_num
)
1561 dev
->real_num_tx_queues
= txq
;
1562 else if (txq
< real_num
) {
1563 dev
->real_num_tx_queues
= txq
;
1564 qdisc_reset_all_tx_gt(dev
, txq
);
1567 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
1569 static inline void __netif_reschedule(struct Qdisc
*q
)
1571 struct softnet_data
*sd
;
1572 unsigned long flags
;
1574 local_irq_save(flags
);
1575 sd
= &__get_cpu_var(softnet_data
);
1576 q
->next_sched
= NULL
;
1577 *sd
->output_queue_tailp
= q
;
1578 sd
->output_queue_tailp
= &q
->next_sched
;
1579 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1580 local_irq_restore(flags
);
1583 void __netif_schedule(struct Qdisc
*q
)
1585 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1586 __netif_reschedule(q
);
1588 EXPORT_SYMBOL(__netif_schedule
);
1590 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1592 if (atomic_dec_and_test(&skb
->users
)) {
1593 struct softnet_data
*sd
;
1594 unsigned long flags
;
1596 local_irq_save(flags
);
1597 sd
= &__get_cpu_var(softnet_data
);
1598 skb
->next
= sd
->completion_queue
;
1599 sd
->completion_queue
= skb
;
1600 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1601 local_irq_restore(flags
);
1604 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1606 void dev_kfree_skb_any(struct sk_buff
*skb
)
1608 if (in_irq() || irqs_disabled())
1609 dev_kfree_skb_irq(skb
);
1613 EXPORT_SYMBOL(dev_kfree_skb_any
);
1617 * netif_device_detach - mark device as removed
1618 * @dev: network device
1620 * Mark device as removed from system and therefore no longer available.
1622 void netif_device_detach(struct net_device
*dev
)
1624 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1625 netif_running(dev
)) {
1626 netif_tx_stop_all_queues(dev
);
1629 EXPORT_SYMBOL(netif_device_detach
);
1632 * netif_device_attach - mark device as attached
1633 * @dev: network device
1635 * Mark device as attached from system and restart if needed.
1637 void netif_device_attach(struct net_device
*dev
)
1639 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1640 netif_running(dev
)) {
1641 netif_tx_wake_all_queues(dev
);
1642 __netdev_watchdog_up(dev
);
1645 EXPORT_SYMBOL(netif_device_attach
);
1647 static bool can_checksum_protocol(unsigned long features
, __be16 protocol
)
1649 return ((features
& NETIF_F_GEN_CSUM
) ||
1650 ((features
& NETIF_F_IP_CSUM
) &&
1651 protocol
== htons(ETH_P_IP
)) ||
1652 ((features
& NETIF_F_IPV6_CSUM
) &&
1653 protocol
== htons(ETH_P_IPV6
)) ||
1654 ((features
& NETIF_F_FCOE_CRC
) &&
1655 protocol
== htons(ETH_P_FCOE
)));
1658 static bool dev_can_checksum(struct net_device
*dev
, struct sk_buff
*skb
)
1660 if (can_checksum_protocol(dev
->features
, skb
->protocol
))
1663 if (skb
->protocol
== htons(ETH_P_8021Q
)) {
1664 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
1665 if (can_checksum_protocol(dev
->features
& dev
->vlan_features
,
1666 veh
->h_vlan_encapsulated_proto
))
1674 * skb_dev_set -- assign a new device to a buffer
1675 * @skb: buffer for the new device
1676 * @dev: network device
1678 * If an skb is owned by a device already, we have to reset
1679 * all data private to the namespace a device belongs to
1680 * before assigning it a new device.
1682 #ifdef CONFIG_NET_NS
1683 void skb_set_dev(struct sk_buff
*skb
, struct net_device
*dev
)
1686 if (skb
->dev
&& !net_eq(dev_net(skb
->dev
), dev_net(dev
))) {
1689 skb_init_secmark(skb
);
1693 skb
->ipvs_property
= 0;
1694 #ifdef CONFIG_NET_SCHED
1700 EXPORT_SYMBOL(skb_set_dev
);
1701 #endif /* CONFIG_NET_NS */
1704 * Invalidate hardware checksum when packet is to be mangled, and
1705 * complete checksum manually on outgoing path.
1707 int skb_checksum_help(struct sk_buff
*skb
)
1710 int ret
= 0, offset
;
1712 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1713 goto out_set_summed
;
1715 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1716 /* Let GSO fix up the checksum. */
1717 goto out_set_summed
;
1720 offset
= skb
->csum_start
- skb_headroom(skb
);
1721 BUG_ON(offset
>= skb_headlen(skb
));
1722 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1724 offset
+= skb
->csum_offset
;
1725 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1727 if (skb_cloned(skb
) &&
1728 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1729 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1734 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1736 skb
->ip_summed
= CHECKSUM_NONE
;
1740 EXPORT_SYMBOL(skb_checksum_help
);
1743 * skb_gso_segment - Perform segmentation on skb.
1744 * @skb: buffer to segment
1745 * @features: features for the output path (see dev->features)
1747 * This function segments the given skb and returns a list of segments.
1749 * It may return NULL if the skb requires no segmentation. This is
1750 * only possible when GSO is used for verifying header integrity.
1752 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, int features
)
1754 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1755 struct packet_type
*ptype
;
1756 __be16 type
= skb
->protocol
;
1759 skb_reset_mac_header(skb
);
1760 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1761 __skb_pull(skb
, skb
->mac_len
);
1763 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1764 struct net_device
*dev
= skb
->dev
;
1765 struct ethtool_drvinfo info
= {};
1767 if (dev
&& dev
->ethtool_ops
&& dev
->ethtool_ops
->get_drvinfo
)
1768 dev
->ethtool_ops
->get_drvinfo(dev
, &info
);
1770 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1772 info
.driver
, dev
? dev
->features
: 0L,
1773 skb
->sk
? skb
->sk
->sk_route_caps
: 0L,
1774 skb
->len
, skb
->data_len
, skb
->ip_summed
);
1776 if (skb_header_cloned(skb
) &&
1777 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1778 return ERR_PTR(err
);
1782 list_for_each_entry_rcu(ptype
,
1783 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1784 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1785 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1786 err
= ptype
->gso_send_check(skb
);
1787 segs
= ERR_PTR(err
);
1788 if (err
|| skb_gso_ok(skb
, features
))
1790 __skb_push(skb
, (skb
->data
-
1791 skb_network_header(skb
)));
1793 segs
= ptype
->gso_segment(skb
, features
);
1799 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1803 EXPORT_SYMBOL(skb_gso_segment
);
1805 /* Take action when hardware reception checksum errors are detected. */
1807 void netdev_rx_csum_fault(struct net_device
*dev
)
1809 if (net_ratelimit()) {
1810 printk(KERN_ERR
"%s: hw csum failure.\n",
1811 dev
? dev
->name
: "<unknown>");
1815 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1818 /* Actually, we should eliminate this check as soon as we know, that:
1819 * 1. IOMMU is present and allows to map all the memory.
1820 * 2. No high memory really exists on this machine.
1823 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1825 #ifdef CONFIG_HIGHMEM
1827 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
1828 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1829 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1833 if (PCI_DMA_BUS_IS_PHYS
) {
1834 struct device
*pdev
= dev
->dev
.parent
;
1838 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1839 dma_addr_t addr
= page_to_phys(skb_shinfo(skb
)->frags
[i
].page
);
1840 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
1849 void (*destructor
)(struct sk_buff
*skb
);
1852 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1854 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1856 struct dev_gso_cb
*cb
;
1859 struct sk_buff
*nskb
= skb
->next
;
1861 skb
->next
= nskb
->next
;
1864 } while (skb
->next
);
1866 cb
= DEV_GSO_CB(skb
);
1868 cb
->destructor(skb
);
1872 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1873 * @skb: buffer to segment
1875 * This function segments the given skb and stores the list of segments
1878 static int dev_gso_segment(struct sk_buff
*skb
)
1880 struct net_device
*dev
= skb
->dev
;
1881 struct sk_buff
*segs
;
1882 int features
= dev
->features
& ~(illegal_highdma(dev
, skb
) ?
1885 segs
= skb_gso_segment(skb
, features
);
1887 /* Verifying header integrity only. */
1892 return PTR_ERR(segs
);
1895 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
1896 skb
->destructor
= dev_gso_skb_destructor
;
1902 * Try to orphan skb early, right before transmission by the device.
1903 * We cannot orphan skb if tx timestamp is requested or the sk-reference
1904 * is needed on driver level for other reasons, e.g. see net/can/raw.c
1906 static inline void skb_orphan_try(struct sk_buff
*skb
)
1908 struct sock
*sk
= skb
->sk
;
1910 if (sk
&& !skb_shinfo(skb
)->tx_flags
) {
1911 /* skb_tx_hash() wont be able to get sk.
1912 * We copy sk_hash into skb->rxhash
1915 skb
->rxhash
= sk
->sk_hash
;
1921 * Returns true if either:
1922 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
1923 * 2. skb is fragmented and the device does not support SG, or if
1924 * at least one of fragments is in highmem and device does not
1925 * support DMA from it.
1927 static inline int skb_needs_linearize(struct sk_buff
*skb
,
1928 struct net_device
*dev
)
1930 return skb_is_nonlinear(skb
) &&
1931 ((skb_has_frag_list(skb
) && !(dev
->features
& NETIF_F_FRAGLIST
)) ||
1932 (skb_shinfo(skb
)->nr_frags
&& (!(dev
->features
& NETIF_F_SG
) ||
1933 illegal_highdma(dev
, skb
))));
1936 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
1937 struct netdev_queue
*txq
)
1939 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1940 int rc
= NETDEV_TX_OK
;
1942 if (likely(!skb
->next
)) {
1943 if (!list_empty(&ptype_all
))
1944 dev_queue_xmit_nit(skb
, dev
);
1947 * If device doesnt need skb->dst, release it right now while
1948 * its hot in this cpu cache
1950 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
1953 skb_orphan_try(skb
);
1955 if (netif_needs_gso(dev
, skb
)) {
1956 if (unlikely(dev_gso_segment(skb
)))
1961 if (skb_needs_linearize(skb
, dev
) &&
1962 __skb_linearize(skb
))
1965 /* If packet is not checksummed and device does not
1966 * support checksumming for this protocol, complete
1967 * checksumming here.
1969 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1970 skb_set_transport_header(skb
, skb
->csum_start
-
1972 if (!dev_can_checksum(dev
, skb
) &&
1973 skb_checksum_help(skb
))
1978 rc
= ops
->ndo_start_xmit(skb
, dev
);
1979 if (rc
== NETDEV_TX_OK
)
1980 txq_trans_update(txq
);
1986 struct sk_buff
*nskb
= skb
->next
;
1988 skb
->next
= nskb
->next
;
1992 * If device doesnt need nskb->dst, release it right now while
1993 * its hot in this cpu cache
1995 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
1998 rc
= ops
->ndo_start_xmit(nskb
, dev
);
1999 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2000 if (rc
& ~NETDEV_TX_MASK
)
2001 goto out_kfree_gso_skb
;
2002 nskb
->next
= skb
->next
;
2006 txq_trans_update(txq
);
2007 if (unlikely(netif_tx_queue_stopped(txq
) && skb
->next
))
2008 return NETDEV_TX_BUSY
;
2009 } while (skb
->next
);
2012 if (likely(skb
->next
== NULL
))
2013 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2019 static u32 hashrnd __read_mostly
;
2021 u16
skb_tx_hash(const struct net_device
*dev
, const struct sk_buff
*skb
)
2025 if (skb_rx_queue_recorded(skb
)) {
2026 hash
= skb_get_rx_queue(skb
);
2027 while (unlikely(hash
>= dev
->real_num_tx_queues
))
2028 hash
-= dev
->real_num_tx_queues
;
2032 if (skb
->sk
&& skb
->sk
->sk_hash
)
2033 hash
= skb
->sk
->sk_hash
;
2035 hash
= (__force u16
) skb
->protocol
^ skb
->rxhash
;
2036 hash
= jhash_1word(hash
, hashrnd
);
2038 return (u16
) (((u64
) hash
* dev
->real_num_tx_queues
) >> 32);
2040 EXPORT_SYMBOL(skb_tx_hash
);
2042 static inline u16
dev_cap_txqueue(struct net_device
*dev
, u16 queue_index
)
2044 if (unlikely(queue_index
>= dev
->real_num_tx_queues
)) {
2045 if (net_ratelimit()) {
2046 pr_warning("%s selects TX queue %d, but "
2047 "real number of TX queues is %d\n",
2048 dev
->name
, queue_index
, dev
->real_num_tx_queues
);
2055 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
2056 struct sk_buff
*skb
)
2059 struct sock
*sk
= skb
->sk
;
2061 queue_index
= sk_tx_queue_get(sk
);
2062 if (queue_index
< 0) {
2063 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2065 if (ops
->ndo_select_queue
) {
2066 queue_index
= ops
->ndo_select_queue(dev
, skb
);
2067 queue_index
= dev_cap_txqueue(dev
, queue_index
);
2070 if (dev
->real_num_tx_queues
> 1)
2071 queue_index
= skb_tx_hash(dev
, skb
);
2074 struct dst_entry
*dst
= rcu_dereference_check(sk
->sk_dst_cache
, 1);
2076 if (dst
&& skb_dst(skb
) == dst
)
2077 sk_tx_queue_set(sk
, queue_index
);
2082 skb_set_queue_mapping(skb
, queue_index
);
2083 return netdev_get_tx_queue(dev
, queue_index
);
2086 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2087 struct net_device
*dev
,
2088 struct netdev_queue
*txq
)
2090 spinlock_t
*root_lock
= qdisc_lock(q
);
2091 bool contended
= qdisc_is_running(q
);
2095 * Heuristic to force contended enqueues to serialize on a
2096 * separate lock before trying to get qdisc main lock.
2097 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2098 * and dequeue packets faster.
2100 if (unlikely(contended
))
2101 spin_lock(&q
->busylock
);
2103 spin_lock(root_lock
);
2104 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2107 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2108 qdisc_run_begin(q
)) {
2110 * This is a work-conserving queue; there are no old skbs
2111 * waiting to be sent out; and the qdisc is not running -
2112 * xmit the skb directly.
2114 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2116 __qdisc_update_bstats(q
, skb
->len
);
2117 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2118 if (unlikely(contended
)) {
2119 spin_unlock(&q
->busylock
);
2126 rc
= NET_XMIT_SUCCESS
;
2129 rc
= qdisc_enqueue_root(skb
, q
);
2130 if (qdisc_run_begin(q
)) {
2131 if (unlikely(contended
)) {
2132 spin_unlock(&q
->busylock
);
2138 spin_unlock(root_lock
);
2139 if (unlikely(contended
))
2140 spin_unlock(&q
->busylock
);
2145 * dev_queue_xmit - transmit a buffer
2146 * @skb: buffer to transmit
2148 * Queue a buffer for transmission to a network device. The caller must
2149 * have set the device and priority and built the buffer before calling
2150 * this function. The function can be called from an interrupt.
2152 * A negative errno code is returned on a failure. A success does not
2153 * guarantee the frame will be transmitted as it may be dropped due
2154 * to congestion or traffic shaping.
2156 * -----------------------------------------------------------------------------------
2157 * I notice this method can also return errors from the queue disciplines,
2158 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2161 * Regardless of the return value, the skb is consumed, so it is currently
2162 * difficult to retry a send to this method. (You can bump the ref count
2163 * before sending to hold a reference for retry if you are careful.)
2165 * When calling this method, interrupts MUST be enabled. This is because
2166 * the BH enable code must have IRQs enabled so that it will not deadlock.
2169 int dev_queue_xmit(struct sk_buff
*skb
)
2171 struct net_device
*dev
= skb
->dev
;
2172 struct netdev_queue
*txq
;
2176 /* Disable soft irqs for various locks below. Also
2177 * stops preemption for RCU.
2181 txq
= dev_pick_tx(dev
, skb
);
2182 q
= rcu_dereference_bh(txq
->qdisc
);
2184 #ifdef CONFIG_NET_CLS_ACT
2185 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2188 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2192 /* The device has no queue. Common case for software devices:
2193 loopback, all the sorts of tunnels...
2195 Really, it is unlikely that netif_tx_lock protection is necessary
2196 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2198 However, it is possible, that they rely on protection
2201 Check this and shot the lock. It is not prone from deadlocks.
2202 Either shot noqueue qdisc, it is even simpler 8)
2204 if (dev
->flags
& IFF_UP
) {
2205 int cpu
= smp_processor_id(); /* ok because BHs are off */
2207 if (txq
->xmit_lock_owner
!= cpu
) {
2209 HARD_TX_LOCK(dev
, txq
, cpu
);
2211 if (!netif_tx_queue_stopped(txq
)) {
2212 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2213 if (dev_xmit_complete(rc
)) {
2214 HARD_TX_UNLOCK(dev
, txq
);
2218 HARD_TX_UNLOCK(dev
, txq
);
2219 if (net_ratelimit())
2220 printk(KERN_CRIT
"Virtual device %s asks to "
2221 "queue packet!\n", dev
->name
);
2223 /* Recursion is detected! It is possible,
2225 if (net_ratelimit())
2226 printk(KERN_CRIT
"Dead loop on virtual device "
2227 "%s, fix it urgently!\n", dev
->name
);
2232 rcu_read_unlock_bh();
2237 rcu_read_unlock_bh();
2240 EXPORT_SYMBOL(dev_queue_xmit
);
2243 /*=======================================================================
2245 =======================================================================*/
2247 int netdev_max_backlog __read_mostly
= 1000;
2248 int netdev_tstamp_prequeue __read_mostly
= 1;
2249 int netdev_budget __read_mostly
= 300;
2250 int weight_p __read_mostly
= 64; /* old backlog weight */
2252 /* Called with irq disabled */
2253 static inline void ____napi_schedule(struct softnet_data
*sd
,
2254 struct napi_struct
*napi
)
2256 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2257 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2261 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2262 * and src/dst port numbers. Returns a non-zero hash number on success
2265 __u32
__skb_get_rxhash(struct sk_buff
*skb
)
2267 int nhoff
, hash
= 0, poff
;
2268 struct ipv6hdr
*ip6
;
2271 u32 addr1
, addr2
, ihl
;
2277 nhoff
= skb_network_offset(skb
);
2279 switch (skb
->protocol
) {
2280 case __constant_htons(ETH_P_IP
):
2281 if (!pskb_may_pull(skb
, sizeof(*ip
) + nhoff
))
2284 ip
= (struct iphdr
*) (skb
->data
+ nhoff
);
2285 if (ip
->frag_off
& htons(IP_MF
| IP_OFFSET
))
2288 ip_proto
= ip
->protocol
;
2289 addr1
= (__force u32
) ip
->saddr
;
2290 addr2
= (__force u32
) ip
->daddr
;
2293 case __constant_htons(ETH_P_IPV6
):
2294 if (!pskb_may_pull(skb
, sizeof(*ip6
) + nhoff
))
2297 ip6
= (struct ipv6hdr
*) (skb
->data
+ nhoff
);
2298 ip_proto
= ip6
->nexthdr
;
2299 addr1
= (__force u32
) ip6
->saddr
.s6_addr32
[3];
2300 addr2
= (__force u32
) ip6
->daddr
.s6_addr32
[3];
2308 poff
= proto_ports_offset(ip_proto
);
2310 nhoff
+= ihl
* 4 + poff
;
2311 if (pskb_may_pull(skb
, nhoff
+ 4)) {
2312 ports
.v32
= * (__force u32
*) (skb
->data
+ nhoff
);
2313 if (ports
.v16
[1] < ports
.v16
[0])
2314 swap(ports
.v16
[0], ports
.v16
[1]);
2318 /* get a consistent hash (same value on both flow directions) */
2322 hash
= jhash_3words(addr1
, addr2
, ports
.v32
, hashrnd
);
2329 EXPORT_SYMBOL(__skb_get_rxhash
);
2333 /* One global table that all flow-based protocols share. */
2334 struct rps_sock_flow_table
*rps_sock_flow_table __read_mostly
;
2335 EXPORT_SYMBOL(rps_sock_flow_table
);
2338 * get_rps_cpu is called from netif_receive_skb and returns the target
2339 * CPU from the RPS map of the receiving queue for a given skb.
2340 * rcu_read_lock must be held on entry.
2342 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2343 struct rps_dev_flow
**rflowp
)
2345 struct netdev_rx_queue
*rxqueue
;
2346 struct rps_map
*map
= NULL
;
2347 struct rps_dev_flow_table
*flow_table
;
2348 struct rps_sock_flow_table
*sock_flow_table
;
2352 if (skb_rx_queue_recorded(skb
)) {
2353 u16 index
= skb_get_rx_queue(skb
);
2354 if (unlikely(index
>= dev
->num_rx_queues
)) {
2355 WARN_ONCE(dev
->num_rx_queues
> 1, "%s received packet "
2356 "on queue %u, but number of RX queues is %u\n",
2357 dev
->name
, index
, dev
->num_rx_queues
);
2360 rxqueue
= dev
->_rx
+ index
;
2364 if (rxqueue
->rps_map
) {
2365 map
= rcu_dereference(rxqueue
->rps_map
);
2366 if (map
&& map
->len
== 1) {
2367 tcpu
= map
->cpus
[0];
2368 if (cpu_online(tcpu
))
2372 } else if (!rxqueue
->rps_flow_table
) {
2376 skb_reset_network_header(skb
);
2377 if (!skb_get_rxhash(skb
))
2380 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2381 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
2382 if (flow_table
&& sock_flow_table
) {
2384 struct rps_dev_flow
*rflow
;
2386 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
2389 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
2390 sock_flow_table
->mask
];
2393 * If the desired CPU (where last recvmsg was done) is
2394 * different from current CPU (one in the rx-queue flow
2395 * table entry), switch if one of the following holds:
2396 * - Current CPU is unset (equal to RPS_NO_CPU).
2397 * - Current CPU is offline.
2398 * - The current CPU's queue tail has advanced beyond the
2399 * last packet that was enqueued using this table entry.
2400 * This guarantees that all previous packets for the flow
2401 * have been dequeued, thus preserving in order delivery.
2403 if (unlikely(tcpu
!= next_cpu
) &&
2404 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
2405 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
2406 rflow
->last_qtail
)) >= 0)) {
2407 tcpu
= rflow
->cpu
= next_cpu
;
2408 if (tcpu
!= RPS_NO_CPU
)
2409 rflow
->last_qtail
= per_cpu(softnet_data
,
2410 tcpu
).input_queue_head
;
2412 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
2420 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2422 if (cpu_online(tcpu
)) {
2432 /* Called from hardirq (IPI) context */
2433 static void rps_trigger_softirq(void *data
)
2435 struct softnet_data
*sd
= data
;
2437 ____napi_schedule(sd
, &sd
->backlog
);
2441 #endif /* CONFIG_RPS */
2444 * Check if this softnet_data structure is another cpu one
2445 * If yes, queue it to our IPI list and return 1
2448 static int rps_ipi_queued(struct softnet_data
*sd
)
2451 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
2454 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
2455 mysd
->rps_ipi_list
= sd
;
2457 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2460 #endif /* CONFIG_RPS */
2465 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2466 * queue (may be a remote CPU queue).
2468 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
2469 unsigned int *qtail
)
2471 struct softnet_data
*sd
;
2472 unsigned long flags
;
2474 sd
= &per_cpu(softnet_data
, cpu
);
2476 local_irq_save(flags
);
2479 if (skb_queue_len(&sd
->input_pkt_queue
) <= netdev_max_backlog
) {
2480 if (skb_queue_len(&sd
->input_pkt_queue
)) {
2482 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
2483 input_queue_tail_incr_save(sd
, qtail
);
2485 local_irq_restore(flags
);
2486 return NET_RX_SUCCESS
;
2489 /* Schedule NAPI for backlog device
2490 * We can use non atomic operation since we own the queue lock
2492 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
2493 if (!rps_ipi_queued(sd
))
2494 ____napi_schedule(sd
, &sd
->backlog
);
2502 local_irq_restore(flags
);
2509 * netif_rx - post buffer to the network code
2510 * @skb: buffer to post
2512 * This function receives a packet from a device driver and queues it for
2513 * the upper (protocol) levels to process. It always succeeds. The buffer
2514 * may be dropped during processing for congestion control or by the
2518 * NET_RX_SUCCESS (no congestion)
2519 * NET_RX_DROP (packet was dropped)
2523 int netif_rx(struct sk_buff
*skb
)
2527 /* if netpoll wants it, pretend we never saw it */
2528 if (netpoll_rx(skb
))
2531 if (netdev_tstamp_prequeue
)
2532 net_timestamp_check(skb
);
2536 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2542 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2544 cpu
= smp_processor_id();
2546 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
2554 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
2560 EXPORT_SYMBOL(netif_rx
);
2562 int netif_rx_ni(struct sk_buff
*skb
)
2567 err
= netif_rx(skb
);
2568 if (local_softirq_pending())
2574 EXPORT_SYMBOL(netif_rx_ni
);
2576 static void net_tx_action(struct softirq_action
*h
)
2578 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2580 if (sd
->completion_queue
) {
2581 struct sk_buff
*clist
;
2583 local_irq_disable();
2584 clist
= sd
->completion_queue
;
2585 sd
->completion_queue
= NULL
;
2589 struct sk_buff
*skb
= clist
;
2590 clist
= clist
->next
;
2592 WARN_ON(atomic_read(&skb
->users
));
2597 if (sd
->output_queue
) {
2600 local_irq_disable();
2601 head
= sd
->output_queue
;
2602 sd
->output_queue
= NULL
;
2603 sd
->output_queue_tailp
= &sd
->output_queue
;
2607 struct Qdisc
*q
= head
;
2608 spinlock_t
*root_lock
;
2610 head
= head
->next_sched
;
2612 root_lock
= qdisc_lock(q
);
2613 if (spin_trylock(root_lock
)) {
2614 smp_mb__before_clear_bit();
2615 clear_bit(__QDISC_STATE_SCHED
,
2618 spin_unlock(root_lock
);
2620 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
2622 __netif_reschedule(q
);
2624 smp_mb__before_clear_bit();
2625 clear_bit(__QDISC_STATE_SCHED
,
2633 static inline int deliver_skb(struct sk_buff
*skb
,
2634 struct packet_type
*pt_prev
,
2635 struct net_device
*orig_dev
)
2637 atomic_inc(&skb
->users
);
2638 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2641 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2642 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2643 /* This hook is defined here for ATM LANE */
2644 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
2645 unsigned char *addr
) __read_mostly
;
2646 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
2649 #ifdef CONFIG_NET_CLS_ACT
2650 /* TODO: Maybe we should just force sch_ingress to be compiled in
2651 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2652 * a compare and 2 stores extra right now if we dont have it on
2653 * but have CONFIG_NET_CLS_ACT
2654 * NOTE: This doesnt stop any functionality; if you dont have
2655 * the ingress scheduler, you just cant add policies on ingress.
2658 static int ing_filter(struct sk_buff
*skb
)
2660 struct net_device
*dev
= skb
->dev
;
2661 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
2662 struct netdev_queue
*rxq
;
2663 int result
= TC_ACT_OK
;
2666 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
2667 if (net_ratelimit())
2668 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2669 skb
->skb_iif
, dev
->ifindex
);
2673 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
2674 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
2676 rxq
= &dev
->rx_queue
;
2679 if (q
!= &noop_qdisc
) {
2680 spin_lock(qdisc_lock(q
));
2681 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
2682 result
= qdisc_enqueue_root(skb
, q
);
2683 spin_unlock(qdisc_lock(q
));
2689 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
2690 struct packet_type
**pt_prev
,
2691 int *ret
, struct net_device
*orig_dev
)
2693 if (skb
->dev
->rx_queue
.qdisc
== &noop_qdisc
)
2697 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2701 switch (ing_filter(skb
)) {
2715 * netif_nit_deliver - deliver received packets to network taps
2718 * This function is used to deliver incoming packets to network
2719 * taps. It should be used when the normal netif_receive_skb path
2720 * is bypassed, for example because of VLAN acceleration.
2722 void netif_nit_deliver(struct sk_buff
*skb
)
2724 struct packet_type
*ptype
;
2726 if (list_empty(&ptype_all
))
2729 skb_reset_network_header(skb
);
2730 skb_reset_transport_header(skb
);
2731 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2734 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2735 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
)
2736 deliver_skb(skb
, ptype
, skb
->dev
);
2742 * netdev_rx_handler_register - register receive handler
2743 * @dev: device to register a handler for
2744 * @rx_handler: receive handler to register
2745 * @rx_handler_data: data pointer that is used by rx handler
2747 * Register a receive hander for a device. This handler will then be
2748 * called from __netif_receive_skb. A negative errno code is returned
2751 * The caller must hold the rtnl_mutex.
2753 int netdev_rx_handler_register(struct net_device
*dev
,
2754 rx_handler_func_t
*rx_handler
,
2755 void *rx_handler_data
)
2759 if (dev
->rx_handler
)
2762 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
2763 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
2767 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
2770 * netdev_rx_handler_unregister - unregister receive handler
2771 * @dev: device to unregister a handler from
2773 * Unregister a receive hander from a device.
2775 * The caller must hold the rtnl_mutex.
2777 void netdev_rx_handler_unregister(struct net_device
*dev
)
2781 rcu_assign_pointer(dev
->rx_handler
, NULL
);
2782 rcu_assign_pointer(dev
->rx_handler_data
, NULL
);
2784 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
2786 static inline void skb_bond_set_mac_by_master(struct sk_buff
*skb
,
2787 struct net_device
*master
)
2789 if (skb
->pkt_type
== PACKET_HOST
) {
2790 u16
*dest
= (u16
*) eth_hdr(skb
)->h_dest
;
2792 memcpy(dest
, master
->dev_addr
, ETH_ALEN
);
2796 /* On bonding slaves other than the currently active slave, suppress
2797 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2798 * ARP on active-backup slaves with arp_validate enabled.
2800 int __skb_bond_should_drop(struct sk_buff
*skb
, struct net_device
*master
)
2802 struct net_device
*dev
= skb
->dev
;
2804 if (master
->priv_flags
& IFF_MASTER_ARPMON
)
2805 dev
->last_rx
= jiffies
;
2807 if ((master
->priv_flags
& IFF_MASTER_ALB
) &&
2808 (master
->priv_flags
& IFF_BRIDGE_PORT
)) {
2809 /* Do address unmangle. The local destination address
2810 * will be always the one master has. Provides the right
2811 * functionality in a bridge.
2813 skb_bond_set_mac_by_master(skb
, master
);
2816 if (dev
->priv_flags
& IFF_SLAVE_INACTIVE
) {
2817 if ((dev
->priv_flags
& IFF_SLAVE_NEEDARP
) &&
2818 skb
->protocol
== __cpu_to_be16(ETH_P_ARP
))
2821 if (master
->priv_flags
& IFF_MASTER_ALB
) {
2822 if (skb
->pkt_type
!= PACKET_BROADCAST
&&
2823 skb
->pkt_type
!= PACKET_MULTICAST
)
2826 if (master
->priv_flags
& IFF_MASTER_8023AD
&&
2827 skb
->protocol
== __cpu_to_be16(ETH_P_SLOW
))
2834 EXPORT_SYMBOL(__skb_bond_should_drop
);
2836 static int __netif_receive_skb(struct sk_buff
*skb
)
2838 struct packet_type
*ptype
, *pt_prev
;
2839 rx_handler_func_t
*rx_handler
;
2840 struct net_device
*orig_dev
;
2841 struct net_device
*master
;
2842 struct net_device
*null_or_orig
;
2843 struct net_device
*orig_or_bond
;
2844 int ret
= NET_RX_DROP
;
2847 if (!netdev_tstamp_prequeue
)
2848 net_timestamp_check(skb
);
2850 if (vlan_tx_tag_present(skb
))
2851 vlan_hwaccel_do_receive(skb
);
2853 /* if we've gotten here through NAPI, check netpoll */
2854 if (netpoll_receive_skb(skb
))
2858 skb
->skb_iif
= skb
->dev
->ifindex
;
2861 * bonding note: skbs received on inactive slaves should only
2862 * be delivered to pkt handlers that are exact matches. Also
2863 * the deliver_no_wcard flag will be set. If packet handlers
2864 * are sensitive to duplicate packets these skbs will need to
2865 * be dropped at the handler. The vlan accel path may have
2866 * already set the deliver_no_wcard flag.
2868 null_or_orig
= NULL
;
2869 orig_dev
= skb
->dev
;
2870 master
= ACCESS_ONCE(orig_dev
->master
);
2871 if (skb
->deliver_no_wcard
)
2872 null_or_orig
= orig_dev
;
2874 if (skb_bond_should_drop(skb
, master
)) {
2875 skb
->deliver_no_wcard
= 1;
2876 null_or_orig
= orig_dev
; /* deliver only exact match */
2881 __this_cpu_inc(softnet_data
.processed
);
2882 skb_reset_network_header(skb
);
2883 skb_reset_transport_header(skb
);
2884 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2890 #ifdef CONFIG_NET_CLS_ACT
2891 if (skb
->tc_verd
& TC_NCLS
) {
2892 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
2897 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2898 if (ptype
->dev
== null_or_orig
|| ptype
->dev
== skb
->dev
||
2899 ptype
->dev
== orig_dev
) {
2901 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2906 #ifdef CONFIG_NET_CLS_ACT
2907 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
2913 /* Handle special case of bridge or macvlan */
2914 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
2917 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2920 skb
= rx_handler(skb
);
2926 * Make sure frames received on VLAN interfaces stacked on
2927 * bonding interfaces still make their way to any base bonding
2928 * device that may have registered for a specific ptype. The
2929 * handler may have to adjust skb->dev and orig_dev.
2931 orig_or_bond
= orig_dev
;
2932 if ((skb
->dev
->priv_flags
& IFF_802_1Q_VLAN
) &&
2933 (vlan_dev_real_dev(skb
->dev
)->priv_flags
& IFF_BONDING
)) {
2934 orig_or_bond
= vlan_dev_real_dev(skb
->dev
);
2937 type
= skb
->protocol
;
2938 list_for_each_entry_rcu(ptype
,
2939 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
2940 if (ptype
->type
== type
&& (ptype
->dev
== null_or_orig
||
2941 ptype
->dev
== skb
->dev
|| ptype
->dev
== orig_dev
||
2942 ptype
->dev
== orig_or_bond
)) {
2944 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2950 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2953 /* Jamal, now you will not able to escape explaining
2954 * me how you were going to use this. :-)
2965 * netif_receive_skb - process receive buffer from network
2966 * @skb: buffer to process
2968 * netif_receive_skb() is the main receive data processing function.
2969 * It always succeeds. The buffer may be dropped during processing
2970 * for congestion control or by the protocol layers.
2972 * This function may only be called from softirq context and interrupts
2973 * should be enabled.
2975 * Return values (usually ignored):
2976 * NET_RX_SUCCESS: no congestion
2977 * NET_RX_DROP: packet was dropped
2979 int netif_receive_skb(struct sk_buff
*skb
)
2981 if (netdev_tstamp_prequeue
)
2982 net_timestamp_check(skb
);
2984 if (skb_defer_rx_timestamp(skb
))
2985 return NET_RX_SUCCESS
;
2989 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2994 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2997 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3001 ret
= __netif_receive_skb(skb
);
3007 return __netif_receive_skb(skb
);
3010 EXPORT_SYMBOL(netif_receive_skb
);
3012 /* Network device is going away, flush any packets still pending
3013 * Called with irqs disabled.
3015 static void flush_backlog(void *arg
)
3017 struct net_device
*dev
= arg
;
3018 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3019 struct sk_buff
*skb
, *tmp
;
3022 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3023 if (skb
->dev
== dev
) {
3024 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3026 input_queue_head_incr(sd
);
3031 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3032 if (skb
->dev
== dev
) {
3033 __skb_unlink(skb
, &sd
->process_queue
);
3035 input_queue_head_incr(sd
);
3040 static int napi_gro_complete(struct sk_buff
*skb
)
3042 struct packet_type
*ptype
;
3043 __be16 type
= skb
->protocol
;
3044 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3047 if (NAPI_GRO_CB(skb
)->count
== 1) {
3048 skb_shinfo(skb
)->gso_size
= 0;
3053 list_for_each_entry_rcu(ptype
, head
, list
) {
3054 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
3057 err
= ptype
->gro_complete(skb
);
3063 WARN_ON(&ptype
->list
== head
);
3065 return NET_RX_SUCCESS
;
3069 return netif_receive_skb(skb
);
3072 inline void napi_gro_flush(struct napi_struct
*napi
)
3074 struct sk_buff
*skb
, *next
;
3076 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3079 napi_gro_complete(skb
);
3082 napi
->gro_count
= 0;
3083 napi
->gro_list
= NULL
;
3085 EXPORT_SYMBOL(napi_gro_flush
);
3087 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3089 struct sk_buff
**pp
= NULL
;
3090 struct packet_type
*ptype
;
3091 __be16 type
= skb
->protocol
;
3092 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3095 enum gro_result ret
;
3097 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3100 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3104 list_for_each_entry_rcu(ptype
, head
, list
) {
3105 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
3108 skb_set_network_header(skb
, skb_gro_offset(skb
));
3109 mac_len
= skb
->network_header
- skb
->mac_header
;
3110 skb
->mac_len
= mac_len
;
3111 NAPI_GRO_CB(skb
)->same_flow
= 0;
3112 NAPI_GRO_CB(skb
)->flush
= 0;
3113 NAPI_GRO_CB(skb
)->free
= 0;
3115 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
3120 if (&ptype
->list
== head
)
3123 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3124 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3127 struct sk_buff
*nskb
= *pp
;
3131 napi_gro_complete(nskb
);
3138 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3142 NAPI_GRO_CB(skb
)->count
= 1;
3143 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3144 skb
->next
= napi
->gro_list
;
3145 napi
->gro_list
= skb
;
3149 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3150 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3152 BUG_ON(skb
->end
- skb
->tail
< grow
);
3154 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3157 skb
->data_len
-= grow
;
3159 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3160 skb_shinfo(skb
)->frags
[0].size
-= grow
;
3162 if (unlikely(!skb_shinfo(skb
)->frags
[0].size
)) {
3163 put_page(skb_shinfo(skb
)->frags
[0].page
);
3164 memmove(skb_shinfo(skb
)->frags
,
3165 skb_shinfo(skb
)->frags
+ 1,
3166 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3177 EXPORT_SYMBOL(dev_gro_receive
);
3179 static inline gro_result_t
3180 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3184 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3185 unsigned long diffs
;
3187 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3188 diffs
|= compare_ether_header(skb_mac_header(p
),
3189 skb_gro_mac_header(skb
));
3190 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3191 NAPI_GRO_CB(p
)->flush
= 0;
3194 return dev_gro_receive(napi
, skb
);
3197 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3201 if (netif_receive_skb(skb
))
3206 case GRO_MERGED_FREE
:
3217 EXPORT_SYMBOL(napi_skb_finish
);
3219 void skb_gro_reset_offset(struct sk_buff
*skb
)
3221 NAPI_GRO_CB(skb
)->data_offset
= 0;
3222 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3223 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3225 if (skb
->mac_header
== skb
->tail
&&
3226 !PageHighMem(skb_shinfo(skb
)->frags
[0].page
)) {
3227 NAPI_GRO_CB(skb
)->frag0
=
3228 page_address(skb_shinfo(skb
)->frags
[0].page
) +
3229 skb_shinfo(skb
)->frags
[0].page_offset
;
3230 NAPI_GRO_CB(skb
)->frag0_len
= skb_shinfo(skb
)->frags
[0].size
;
3233 EXPORT_SYMBOL(skb_gro_reset_offset
);
3235 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3237 skb_gro_reset_offset(skb
);
3239 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
3241 EXPORT_SYMBOL(napi_gro_receive
);
3243 void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3245 __skb_pull(skb
, skb_headlen(skb
));
3246 skb_reserve(skb
, NET_IP_ALIGN
- skb_headroom(skb
));
3250 EXPORT_SYMBOL(napi_reuse_skb
);
3252 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3254 struct sk_buff
*skb
= napi
->skb
;
3257 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3263 EXPORT_SYMBOL(napi_get_frags
);
3265 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3271 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3273 if (ret
== GRO_HELD
)
3274 skb_gro_pull(skb
, -ETH_HLEN
);
3275 else if (netif_receive_skb(skb
))
3280 case GRO_MERGED_FREE
:
3281 napi_reuse_skb(napi
, skb
);
3290 EXPORT_SYMBOL(napi_frags_finish
);
3292 struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3294 struct sk_buff
*skb
= napi
->skb
;
3301 skb_reset_mac_header(skb
);
3302 skb_gro_reset_offset(skb
);
3304 off
= skb_gro_offset(skb
);
3305 hlen
= off
+ sizeof(*eth
);
3306 eth
= skb_gro_header_fast(skb
, off
);
3307 if (skb_gro_header_hard(skb
, hlen
)) {
3308 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3309 if (unlikely(!eth
)) {
3310 napi_reuse_skb(napi
, skb
);
3316 skb_gro_pull(skb
, sizeof(*eth
));
3319 * This works because the only protocols we care about don't require
3320 * special handling. We'll fix it up properly at the end.
3322 skb
->protocol
= eth
->h_proto
;
3327 EXPORT_SYMBOL(napi_frags_skb
);
3329 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3331 struct sk_buff
*skb
= napi_frags_skb(napi
);
3336 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3338 EXPORT_SYMBOL(napi_gro_frags
);
3341 * net_rps_action sends any pending IPI's for rps.
3342 * Note: called with local irq disabled, but exits with local irq enabled.
3344 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
3347 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3350 sd
->rps_ipi_list
= NULL
;
3354 /* Send pending IPI's to kick RPS processing on remote cpus. */
3356 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3358 if (cpu_online(remsd
->cpu
))
3359 __smp_call_function_single(remsd
->cpu
,
3368 static int process_backlog(struct napi_struct
*napi
, int quota
)
3371 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
3374 /* Check if we have pending ipi, its better to send them now,
3375 * not waiting net_rx_action() end.
3377 if (sd
->rps_ipi_list
) {
3378 local_irq_disable();
3379 net_rps_action_and_irq_enable(sd
);
3382 napi
->weight
= weight_p
;
3383 local_irq_disable();
3384 while (work
< quota
) {
3385 struct sk_buff
*skb
;
3388 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
3390 __netif_receive_skb(skb
);
3391 local_irq_disable();
3392 input_queue_head_incr(sd
);
3393 if (++work
>= quota
) {
3400 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3402 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
3403 &sd
->process_queue
);
3405 if (qlen
< quota
- work
) {
3407 * Inline a custom version of __napi_complete().
3408 * only current cpu owns and manipulates this napi,
3409 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3410 * we can use a plain write instead of clear_bit(),
3411 * and we dont need an smp_mb() memory barrier.
3413 list_del(&napi
->poll_list
);
3416 quota
= work
+ qlen
;
3426 * __napi_schedule - schedule for receive
3427 * @n: entry to schedule
3429 * The entry's receive function will be scheduled to run
3431 void __napi_schedule(struct napi_struct
*n
)
3433 unsigned long flags
;
3435 local_irq_save(flags
);
3436 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
3437 local_irq_restore(flags
);
3439 EXPORT_SYMBOL(__napi_schedule
);
3441 void __napi_complete(struct napi_struct
*n
)
3443 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3444 BUG_ON(n
->gro_list
);
3446 list_del(&n
->poll_list
);
3447 smp_mb__before_clear_bit();
3448 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3450 EXPORT_SYMBOL(__napi_complete
);
3452 void napi_complete(struct napi_struct
*n
)
3454 unsigned long flags
;
3457 * don't let napi dequeue from the cpu poll list
3458 * just in case its running on a different cpu
3460 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3464 local_irq_save(flags
);
3466 local_irq_restore(flags
);
3468 EXPORT_SYMBOL(napi_complete
);
3470 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3471 int (*poll
)(struct napi_struct
*, int), int weight
)
3473 INIT_LIST_HEAD(&napi
->poll_list
);
3474 napi
->gro_count
= 0;
3475 napi
->gro_list
= NULL
;
3478 napi
->weight
= weight
;
3479 list_add(&napi
->dev_list
, &dev
->napi_list
);
3481 #ifdef CONFIG_NETPOLL
3482 spin_lock_init(&napi
->poll_lock
);
3483 napi
->poll_owner
= -1;
3485 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3487 EXPORT_SYMBOL(netif_napi_add
);
3489 void netif_napi_del(struct napi_struct
*napi
)
3491 struct sk_buff
*skb
, *next
;
3493 list_del_init(&napi
->dev_list
);
3494 napi_free_frags(napi
);
3496 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3502 napi
->gro_list
= NULL
;
3503 napi
->gro_count
= 0;
3505 EXPORT_SYMBOL(netif_napi_del
);
3507 static void net_rx_action(struct softirq_action
*h
)
3509 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3510 unsigned long time_limit
= jiffies
+ 2;
3511 int budget
= netdev_budget
;
3514 local_irq_disable();
3516 while (!list_empty(&sd
->poll_list
)) {
3517 struct napi_struct
*n
;
3520 /* If softirq window is exhuasted then punt.
3521 * Allow this to run for 2 jiffies since which will allow
3522 * an average latency of 1.5/HZ.
3524 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3529 /* Even though interrupts have been re-enabled, this
3530 * access is safe because interrupts can only add new
3531 * entries to the tail of this list, and only ->poll()
3532 * calls can remove this head entry from the list.
3534 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
3536 have
= netpoll_poll_lock(n
);
3540 /* This NAPI_STATE_SCHED test is for avoiding a race
3541 * with netpoll's poll_napi(). Only the entity which
3542 * obtains the lock and sees NAPI_STATE_SCHED set will
3543 * actually make the ->poll() call. Therefore we avoid
3544 * accidently calling ->poll() when NAPI is not scheduled.
3547 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3548 work
= n
->poll(n
, weight
);
3552 WARN_ON_ONCE(work
> weight
);
3556 local_irq_disable();
3558 /* Drivers must not modify the NAPI state if they
3559 * consume the entire weight. In such cases this code
3560 * still "owns" the NAPI instance and therefore can
3561 * move the instance around on the list at-will.
3563 if (unlikely(work
== weight
)) {
3564 if (unlikely(napi_disable_pending(n
))) {
3567 local_irq_disable();
3569 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
3572 netpoll_poll_unlock(have
);
3575 net_rps_action_and_irq_enable(sd
);
3577 #ifdef CONFIG_NET_DMA
3579 * There may not be any more sk_buffs coming right now, so push
3580 * any pending DMA copies to hardware
3582 dma_issue_pending_all();
3589 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3593 static gifconf_func_t
*gifconf_list
[NPROTO
];
3596 * register_gifconf - register a SIOCGIF handler
3597 * @family: Address family
3598 * @gifconf: Function handler
3600 * Register protocol dependent address dumping routines. The handler
3601 * that is passed must not be freed or reused until it has been replaced
3602 * by another handler.
3604 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
3606 if (family
>= NPROTO
)
3608 gifconf_list
[family
] = gifconf
;
3611 EXPORT_SYMBOL(register_gifconf
);
3615 * Map an interface index to its name (SIOCGIFNAME)
3619 * We need this ioctl for efficient implementation of the
3620 * if_indextoname() function required by the IPv6 API. Without
3621 * it, we would have to search all the interfaces to find a
3625 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
3627 struct net_device
*dev
;
3631 * Fetch the caller's info block.
3634 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3638 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
3644 strcpy(ifr
.ifr_name
, dev
->name
);
3647 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
3653 * Perform a SIOCGIFCONF call. This structure will change
3654 * size eventually, and there is nothing I can do about it.
3655 * Thus we will need a 'compatibility mode'.
3658 static int dev_ifconf(struct net
*net
, char __user
*arg
)
3661 struct net_device
*dev
;
3668 * Fetch the caller's info block.
3671 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
3678 * Loop over the interfaces, and write an info block for each.
3682 for_each_netdev(net
, dev
) {
3683 for (i
= 0; i
< NPROTO
; i
++) {
3684 if (gifconf_list
[i
]) {
3687 done
= gifconf_list
[i
](dev
, NULL
, 0);
3689 done
= gifconf_list
[i
](dev
, pos
+ total
,
3699 * All done. Write the updated control block back to the caller.
3701 ifc
.ifc_len
= total
;
3704 * Both BSD and Solaris return 0 here, so we do too.
3706 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
3709 #ifdef CONFIG_PROC_FS
3711 * This is invoked by the /proc filesystem handler to display a device
3714 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3717 struct net
*net
= seq_file_net(seq
);
3719 struct net_device
*dev
;
3723 return SEQ_START_TOKEN
;
3726 for_each_netdev_rcu(net
, dev
)
3733 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3735 struct net_device
*dev
= (v
== SEQ_START_TOKEN
) ?
3736 first_net_device(seq_file_net(seq
)) :
3737 next_net_device((struct net_device
*)v
);
3740 return rcu_dereference(dev
);
3743 void dev_seq_stop(struct seq_file
*seq
, void *v
)
3749 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
3751 struct rtnl_link_stats64 temp
;
3752 const struct rtnl_link_stats64
*stats
= dev_get_stats(dev
, &temp
);
3754 seq_printf(seq
, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3755 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3756 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
3758 stats
->rx_dropped
+ stats
->rx_missed_errors
,
3759 stats
->rx_fifo_errors
,
3760 stats
->rx_length_errors
+ stats
->rx_over_errors
+
3761 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
3762 stats
->rx_compressed
, stats
->multicast
,
3763 stats
->tx_bytes
, stats
->tx_packets
,
3764 stats
->tx_errors
, stats
->tx_dropped
,
3765 stats
->tx_fifo_errors
, stats
->collisions
,
3766 stats
->tx_carrier_errors
+
3767 stats
->tx_aborted_errors
+
3768 stats
->tx_window_errors
+
3769 stats
->tx_heartbeat_errors
,
3770 stats
->tx_compressed
);
3774 * Called from the PROCfs module. This now uses the new arbitrary sized
3775 * /proc/net interface to create /proc/net/dev
3777 static int dev_seq_show(struct seq_file
*seq
, void *v
)
3779 if (v
== SEQ_START_TOKEN
)
3780 seq_puts(seq
, "Inter-| Receive "
3782 " face |bytes packets errs drop fifo frame "
3783 "compressed multicast|bytes packets errs "
3784 "drop fifo colls carrier compressed\n");
3786 dev_seq_printf_stats(seq
, v
);
3790 static struct softnet_data
*softnet_get_online(loff_t
*pos
)
3792 struct softnet_data
*sd
= NULL
;
3794 while (*pos
< nr_cpu_ids
)
3795 if (cpu_online(*pos
)) {
3796 sd
= &per_cpu(softnet_data
, *pos
);
3803 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3805 return softnet_get_online(pos
);
3808 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3811 return softnet_get_online(pos
);
3814 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
3818 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
3820 struct softnet_data
*sd
= v
;
3822 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3823 sd
->processed
, sd
->dropped
, sd
->time_squeeze
, 0,
3824 0, 0, 0, 0, /* was fastroute */
3825 sd
->cpu_collision
, sd
->received_rps
);
3829 static const struct seq_operations dev_seq_ops
= {
3830 .start
= dev_seq_start
,
3831 .next
= dev_seq_next
,
3832 .stop
= dev_seq_stop
,
3833 .show
= dev_seq_show
,
3836 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
3838 return seq_open_net(inode
, file
, &dev_seq_ops
,
3839 sizeof(struct seq_net_private
));
3842 static const struct file_operations dev_seq_fops
= {
3843 .owner
= THIS_MODULE
,
3844 .open
= dev_seq_open
,
3846 .llseek
= seq_lseek
,
3847 .release
= seq_release_net
,
3850 static const struct seq_operations softnet_seq_ops
= {
3851 .start
= softnet_seq_start
,
3852 .next
= softnet_seq_next
,
3853 .stop
= softnet_seq_stop
,
3854 .show
= softnet_seq_show
,
3857 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
3859 return seq_open(file
, &softnet_seq_ops
);
3862 static const struct file_operations softnet_seq_fops
= {
3863 .owner
= THIS_MODULE
,
3864 .open
= softnet_seq_open
,
3866 .llseek
= seq_lseek
,
3867 .release
= seq_release
,
3870 static void *ptype_get_idx(loff_t pos
)
3872 struct packet_type
*pt
= NULL
;
3876 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
3882 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
3883 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
3892 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3896 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
3899 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3901 struct packet_type
*pt
;
3902 struct list_head
*nxt
;
3906 if (v
== SEQ_START_TOKEN
)
3907 return ptype_get_idx(0);
3910 nxt
= pt
->list
.next
;
3911 if (pt
->type
== htons(ETH_P_ALL
)) {
3912 if (nxt
!= &ptype_all
)
3915 nxt
= ptype_base
[0].next
;
3917 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
3919 while (nxt
== &ptype_base
[hash
]) {
3920 if (++hash
>= PTYPE_HASH_SIZE
)
3922 nxt
= ptype_base
[hash
].next
;
3925 return list_entry(nxt
, struct packet_type
, list
);
3928 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
3934 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
3936 struct packet_type
*pt
= v
;
3938 if (v
== SEQ_START_TOKEN
)
3939 seq_puts(seq
, "Type Device Function\n");
3940 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
3941 if (pt
->type
== htons(ETH_P_ALL
))
3942 seq_puts(seq
, "ALL ");
3944 seq_printf(seq
, "%04x", ntohs(pt
->type
));
3946 seq_printf(seq
, " %-8s %pF\n",
3947 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
3953 static const struct seq_operations ptype_seq_ops
= {
3954 .start
= ptype_seq_start
,
3955 .next
= ptype_seq_next
,
3956 .stop
= ptype_seq_stop
,
3957 .show
= ptype_seq_show
,
3960 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
3962 return seq_open_net(inode
, file
, &ptype_seq_ops
,
3963 sizeof(struct seq_net_private
));
3966 static const struct file_operations ptype_seq_fops
= {
3967 .owner
= THIS_MODULE
,
3968 .open
= ptype_seq_open
,
3970 .llseek
= seq_lseek
,
3971 .release
= seq_release_net
,
3975 static int __net_init
dev_proc_net_init(struct net
*net
)
3979 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
3981 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
3983 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
3986 if (wext_proc_init(net
))
3992 proc_net_remove(net
, "ptype");
3994 proc_net_remove(net
, "softnet_stat");
3996 proc_net_remove(net
, "dev");
4000 static void __net_exit
dev_proc_net_exit(struct net
*net
)
4002 wext_proc_exit(net
);
4004 proc_net_remove(net
, "ptype");
4005 proc_net_remove(net
, "softnet_stat");
4006 proc_net_remove(net
, "dev");
4009 static struct pernet_operations __net_initdata dev_proc_ops
= {
4010 .init
= dev_proc_net_init
,
4011 .exit
= dev_proc_net_exit
,
4014 static int __init
dev_proc_init(void)
4016 return register_pernet_subsys(&dev_proc_ops
);
4019 #define dev_proc_init() 0
4020 #endif /* CONFIG_PROC_FS */
4024 * netdev_set_master - set up master/slave pair
4025 * @slave: slave device
4026 * @master: new master device
4028 * Changes the master device of the slave. Pass %NULL to break the
4029 * bonding. The caller must hold the RTNL semaphore. On a failure
4030 * a negative errno code is returned. On success the reference counts
4031 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4032 * function returns zero.
4034 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
4036 struct net_device
*old
= slave
->master
;
4046 slave
->master
= master
;
4053 slave
->flags
|= IFF_SLAVE
;
4055 slave
->flags
&= ~IFF_SLAVE
;
4057 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
4060 EXPORT_SYMBOL(netdev_set_master
);
4062 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4064 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4066 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4067 ops
->ndo_change_rx_flags(dev
, flags
);
4070 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4072 unsigned short old_flags
= dev
->flags
;
4078 dev
->flags
|= IFF_PROMISC
;
4079 dev
->promiscuity
+= inc
;
4080 if (dev
->promiscuity
== 0) {
4083 * If inc causes overflow, untouch promisc and return error.
4086 dev
->flags
&= ~IFF_PROMISC
;
4088 dev
->promiscuity
-= inc
;
4089 printk(KERN_WARNING
"%s: promiscuity touches roof, "
4090 "set promiscuity failed, promiscuity feature "
4091 "of device might be broken.\n", dev
->name
);
4095 if (dev
->flags
!= old_flags
) {
4096 printk(KERN_INFO
"device %s %s promiscuous mode\n",
4097 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
4099 if (audit_enabled
) {
4100 current_uid_gid(&uid
, &gid
);
4101 audit_log(current
->audit_context
, GFP_ATOMIC
,
4102 AUDIT_ANOM_PROMISCUOUS
,
4103 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4104 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4105 (old_flags
& IFF_PROMISC
),
4106 audit_get_loginuid(current
),
4108 audit_get_sessionid(current
));
4111 dev_change_rx_flags(dev
, IFF_PROMISC
);
4117 * dev_set_promiscuity - update promiscuity count on a device
4121 * Add or remove promiscuity from a device. While the count in the device
4122 * remains above zero the interface remains promiscuous. Once it hits zero
4123 * the device reverts back to normal filtering operation. A negative inc
4124 * value is used to drop promiscuity on the device.
4125 * Return 0 if successful or a negative errno code on error.
4127 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4129 unsigned short old_flags
= dev
->flags
;
4132 err
= __dev_set_promiscuity(dev
, inc
);
4135 if (dev
->flags
!= old_flags
)
4136 dev_set_rx_mode(dev
);
4139 EXPORT_SYMBOL(dev_set_promiscuity
);
4142 * dev_set_allmulti - update allmulti count on a device
4146 * Add or remove reception of all multicast frames to a device. While the
4147 * count in the device remains above zero the interface remains listening
4148 * to all interfaces. Once it hits zero the device reverts back to normal
4149 * filtering operation. A negative @inc value is used to drop the counter
4150 * when releasing a resource needing all multicasts.
4151 * Return 0 if successful or a negative errno code on error.
4154 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4156 unsigned short old_flags
= dev
->flags
;
4160 dev
->flags
|= IFF_ALLMULTI
;
4161 dev
->allmulti
+= inc
;
4162 if (dev
->allmulti
== 0) {
4165 * If inc causes overflow, untouch allmulti and return error.
4168 dev
->flags
&= ~IFF_ALLMULTI
;
4170 dev
->allmulti
-= inc
;
4171 printk(KERN_WARNING
"%s: allmulti touches roof, "
4172 "set allmulti failed, allmulti feature of "
4173 "device might be broken.\n", dev
->name
);
4177 if (dev
->flags
^ old_flags
) {
4178 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4179 dev_set_rx_mode(dev
);
4183 EXPORT_SYMBOL(dev_set_allmulti
);
4186 * Upload unicast and multicast address lists to device and
4187 * configure RX filtering. When the device doesn't support unicast
4188 * filtering it is put in promiscuous mode while unicast addresses
4191 void __dev_set_rx_mode(struct net_device
*dev
)
4193 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4195 /* dev_open will call this function so the list will stay sane. */
4196 if (!(dev
->flags
&IFF_UP
))
4199 if (!netif_device_present(dev
))
4202 if (ops
->ndo_set_rx_mode
)
4203 ops
->ndo_set_rx_mode(dev
);
4205 /* Unicast addresses changes may only happen under the rtnl,
4206 * therefore calling __dev_set_promiscuity here is safe.
4208 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4209 __dev_set_promiscuity(dev
, 1);
4210 dev
->uc_promisc
= 1;
4211 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4212 __dev_set_promiscuity(dev
, -1);
4213 dev
->uc_promisc
= 0;
4216 if (ops
->ndo_set_multicast_list
)
4217 ops
->ndo_set_multicast_list(dev
);
4221 void dev_set_rx_mode(struct net_device
*dev
)
4223 netif_addr_lock_bh(dev
);
4224 __dev_set_rx_mode(dev
);
4225 netif_addr_unlock_bh(dev
);
4229 * dev_get_flags - get flags reported to userspace
4232 * Get the combination of flag bits exported through APIs to userspace.
4234 unsigned dev_get_flags(const struct net_device
*dev
)
4238 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4243 (dev
->gflags
& (IFF_PROMISC
|
4246 if (netif_running(dev
)) {
4247 if (netif_oper_up(dev
))
4248 flags
|= IFF_RUNNING
;
4249 if (netif_carrier_ok(dev
))
4250 flags
|= IFF_LOWER_UP
;
4251 if (netif_dormant(dev
))
4252 flags
|= IFF_DORMANT
;
4257 EXPORT_SYMBOL(dev_get_flags
);
4259 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4261 int old_flags
= dev
->flags
;
4267 * Set the flags on our device.
4270 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4271 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4273 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4277 * Load in the correct multicast list now the flags have changed.
4280 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4281 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4283 dev_set_rx_mode(dev
);
4286 * Have we downed the interface. We handle IFF_UP ourselves
4287 * according to user attempts to set it, rather than blindly
4292 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4293 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4296 dev_set_rx_mode(dev
);
4299 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4300 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4302 dev
->gflags
^= IFF_PROMISC
;
4303 dev_set_promiscuity(dev
, inc
);
4306 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4307 is important. Some (broken) drivers set IFF_PROMISC, when
4308 IFF_ALLMULTI is requested not asking us and not reporting.
4310 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4311 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4313 dev
->gflags
^= IFF_ALLMULTI
;
4314 dev_set_allmulti(dev
, inc
);
4320 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4322 unsigned int changes
= dev
->flags
^ old_flags
;
4324 if (changes
& IFF_UP
) {
4325 if (dev
->flags
& IFF_UP
)
4326 call_netdevice_notifiers(NETDEV_UP
, dev
);
4328 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4331 if (dev
->flags
& IFF_UP
&&
4332 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4333 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4337 * dev_change_flags - change device settings
4339 * @flags: device state flags
4341 * Change settings on device based state flags. The flags are
4342 * in the userspace exported format.
4344 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
4347 int old_flags
= dev
->flags
;
4349 ret
= __dev_change_flags(dev
, flags
);
4353 changes
= old_flags
^ dev
->flags
;
4355 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4357 __dev_notify_flags(dev
, old_flags
);
4360 EXPORT_SYMBOL(dev_change_flags
);
4363 * dev_set_mtu - Change maximum transfer unit
4365 * @new_mtu: new transfer unit
4367 * Change the maximum transfer size of the network device.
4369 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4371 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4374 if (new_mtu
== dev
->mtu
)
4377 /* MTU must be positive. */
4381 if (!netif_device_present(dev
))
4385 if (ops
->ndo_change_mtu
)
4386 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4390 if (!err
&& dev
->flags
& IFF_UP
)
4391 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4394 EXPORT_SYMBOL(dev_set_mtu
);
4397 * dev_set_mac_address - Change Media Access Control Address
4401 * Change the hardware (MAC) address of the device
4403 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4405 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4408 if (!ops
->ndo_set_mac_address
)
4410 if (sa
->sa_family
!= dev
->type
)
4412 if (!netif_device_present(dev
))
4414 err
= ops
->ndo_set_mac_address(dev
, sa
);
4416 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4419 EXPORT_SYMBOL(dev_set_mac_address
);
4422 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4424 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4427 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4433 case SIOCGIFFLAGS
: /* Get interface flags */
4434 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4437 case SIOCGIFMETRIC
: /* Get the metric on the interface
4438 (currently unused) */
4439 ifr
->ifr_metric
= 0;
4442 case SIOCGIFMTU
: /* Get the MTU of a device */
4443 ifr
->ifr_mtu
= dev
->mtu
;
4448 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4450 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4451 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4452 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4460 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4461 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4462 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4463 ifr
->ifr_map
.irq
= dev
->irq
;
4464 ifr
->ifr_map
.dma
= dev
->dma
;
4465 ifr
->ifr_map
.port
= dev
->if_port
;
4469 ifr
->ifr_ifindex
= dev
->ifindex
;
4473 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4477 /* dev_ioctl() should ensure this case
4489 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4491 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4494 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4495 const struct net_device_ops
*ops
;
4500 ops
= dev
->netdev_ops
;
4503 case SIOCSIFFLAGS
: /* Set interface flags */
4504 return dev_change_flags(dev
, ifr
->ifr_flags
);
4506 case SIOCSIFMETRIC
: /* Set the metric on the interface
4507 (currently unused) */
4510 case SIOCSIFMTU
: /* Set the MTU of a device */
4511 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4514 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4516 case SIOCSIFHWBROADCAST
:
4517 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4519 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4520 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4521 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4525 if (ops
->ndo_set_config
) {
4526 if (!netif_device_present(dev
))
4528 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
4533 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4534 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4536 if (!netif_device_present(dev
))
4538 return dev_mc_add_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4541 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4542 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4544 if (!netif_device_present(dev
))
4546 return dev_mc_del_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4549 if (ifr
->ifr_qlen
< 0)
4551 dev
->tx_queue_len
= ifr
->ifr_qlen
;
4555 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
4556 return dev_change_name(dev
, ifr
->ifr_newname
);
4559 * Unknown or private ioctl
4562 if ((cmd
>= SIOCDEVPRIVATE
&&
4563 cmd
<= SIOCDEVPRIVATE
+ 15) ||
4564 cmd
== SIOCBONDENSLAVE
||
4565 cmd
== SIOCBONDRELEASE
||
4566 cmd
== SIOCBONDSETHWADDR
||
4567 cmd
== SIOCBONDSLAVEINFOQUERY
||
4568 cmd
== SIOCBONDINFOQUERY
||
4569 cmd
== SIOCBONDCHANGEACTIVE
||
4570 cmd
== SIOCGMIIPHY
||
4571 cmd
== SIOCGMIIREG
||
4572 cmd
== SIOCSMIIREG
||
4573 cmd
== SIOCBRADDIF
||
4574 cmd
== SIOCBRDELIF
||
4575 cmd
== SIOCSHWTSTAMP
||
4576 cmd
== SIOCWANDEV
) {
4578 if (ops
->ndo_do_ioctl
) {
4579 if (netif_device_present(dev
))
4580 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
4592 * This function handles all "interface"-type I/O control requests. The actual
4593 * 'doing' part of this is dev_ifsioc above.
4597 * dev_ioctl - network device ioctl
4598 * @net: the applicable net namespace
4599 * @cmd: command to issue
4600 * @arg: pointer to a struct ifreq in user space
4602 * Issue ioctl functions to devices. This is normally called by the
4603 * user space syscall interfaces but can sometimes be useful for
4604 * other purposes. The return value is the return from the syscall if
4605 * positive or a negative errno code on error.
4608 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
4614 /* One special case: SIOCGIFCONF takes ifconf argument
4615 and requires shared lock, because it sleeps writing
4619 if (cmd
== SIOCGIFCONF
) {
4621 ret
= dev_ifconf(net
, (char __user
*) arg
);
4625 if (cmd
== SIOCGIFNAME
)
4626 return dev_ifname(net
, (struct ifreq __user
*)arg
);
4628 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
4631 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
4633 colon
= strchr(ifr
.ifr_name
, ':');
4638 * See which interface the caller is talking about.
4643 * These ioctl calls:
4644 * - can be done by all.
4645 * - atomic and do not require locking.
4656 dev_load(net
, ifr
.ifr_name
);
4658 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
4663 if (copy_to_user(arg
, &ifr
,
4664 sizeof(struct ifreq
)))
4670 dev_load(net
, ifr
.ifr_name
);
4672 ret
= dev_ethtool(net
, &ifr
);
4677 if (copy_to_user(arg
, &ifr
,
4678 sizeof(struct ifreq
)))
4684 * These ioctl calls:
4685 * - require superuser power.
4686 * - require strict serialization.
4692 if (!capable(CAP_NET_ADMIN
))
4694 dev_load(net
, ifr
.ifr_name
);
4696 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4701 if (copy_to_user(arg
, &ifr
,
4702 sizeof(struct ifreq
)))
4708 * These ioctl calls:
4709 * - require superuser power.
4710 * - require strict serialization.
4711 * - do not return a value
4721 case SIOCSIFHWBROADCAST
:
4724 case SIOCBONDENSLAVE
:
4725 case SIOCBONDRELEASE
:
4726 case SIOCBONDSETHWADDR
:
4727 case SIOCBONDCHANGEACTIVE
:
4731 if (!capable(CAP_NET_ADMIN
))
4734 case SIOCBONDSLAVEINFOQUERY
:
4735 case SIOCBONDINFOQUERY
:
4736 dev_load(net
, ifr
.ifr_name
);
4738 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4743 /* Get the per device memory space. We can add this but
4744 * currently do not support it */
4746 /* Set the per device memory buffer space.
4747 * Not applicable in our case */
4752 * Unknown or private ioctl.
4755 if (cmd
== SIOCWANDEV
||
4756 (cmd
>= SIOCDEVPRIVATE
&&
4757 cmd
<= SIOCDEVPRIVATE
+ 15)) {
4758 dev_load(net
, ifr
.ifr_name
);
4760 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4762 if (!ret
&& copy_to_user(arg
, &ifr
,
4763 sizeof(struct ifreq
)))
4767 /* Take care of Wireless Extensions */
4768 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
4769 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
4776 * dev_new_index - allocate an ifindex
4777 * @net: the applicable net namespace
4779 * Returns a suitable unique value for a new device interface
4780 * number. The caller must hold the rtnl semaphore or the
4781 * dev_base_lock to be sure it remains unique.
4783 static int dev_new_index(struct net
*net
)
4789 if (!__dev_get_by_index(net
, ifindex
))
4794 /* Delayed registration/unregisteration */
4795 static LIST_HEAD(net_todo_list
);
4797 static void net_set_todo(struct net_device
*dev
)
4799 list_add_tail(&dev
->todo_list
, &net_todo_list
);
4802 static void rollback_registered_many(struct list_head
*head
)
4804 struct net_device
*dev
, *tmp
;
4806 BUG_ON(dev_boot_phase
);
4809 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
4810 /* Some devices call without registering
4811 * for initialization unwind. Remove those
4812 * devices and proceed with the remaining.
4814 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
4815 pr_debug("unregister_netdevice: device %s/%p never "
4816 "was registered\n", dev
->name
, dev
);
4819 list_del(&dev
->unreg_list
);
4823 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
4825 /* If device is running, close it first. */
4828 /* And unlink it from device chain. */
4829 unlist_netdevice(dev
);
4831 dev
->reg_state
= NETREG_UNREGISTERING
;
4836 list_for_each_entry(dev
, head
, unreg_list
) {
4837 /* Shutdown queueing discipline. */
4841 /* Notify protocols, that we are about to destroy
4842 this device. They should clean all the things.
4844 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
4846 if (!dev
->rtnl_link_ops
||
4847 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
4848 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
4851 * Flush the unicast and multicast chains
4856 if (dev
->netdev_ops
->ndo_uninit
)
4857 dev
->netdev_ops
->ndo_uninit(dev
);
4859 /* Notifier chain MUST detach us from master device. */
4860 WARN_ON(dev
->master
);
4862 /* Remove entries from kobject tree */
4863 netdev_unregister_kobject(dev
);
4866 /* Process any work delayed until the end of the batch */
4867 dev
= list_first_entry(head
, struct net_device
, unreg_list
);
4868 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
4872 list_for_each_entry(dev
, head
, unreg_list
)
4876 static void rollback_registered(struct net_device
*dev
)
4880 list_add(&dev
->unreg_list
, &single
);
4881 rollback_registered_many(&single
);
4884 static void __netdev_init_queue_locks_one(struct net_device
*dev
,
4885 struct netdev_queue
*dev_queue
,
4888 spin_lock_init(&dev_queue
->_xmit_lock
);
4889 netdev_set_xmit_lockdep_class(&dev_queue
->_xmit_lock
, dev
->type
);
4890 dev_queue
->xmit_lock_owner
= -1;
4893 static void netdev_init_queue_locks(struct net_device
*dev
)
4895 netdev_for_each_tx_queue(dev
, __netdev_init_queue_locks_one
, NULL
);
4896 __netdev_init_queue_locks_one(dev
, &dev
->rx_queue
, NULL
);
4899 unsigned long netdev_fix_features(unsigned long features
, const char *name
)
4901 /* Fix illegal SG+CSUM combinations. */
4902 if ((features
& NETIF_F_SG
) &&
4903 !(features
& NETIF_F_ALL_CSUM
)) {
4905 printk(KERN_NOTICE
"%s: Dropping NETIF_F_SG since no "
4906 "checksum feature.\n", name
);
4907 features
&= ~NETIF_F_SG
;
4910 /* TSO requires that SG is present as well. */
4911 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_SG
)) {
4913 printk(KERN_NOTICE
"%s: Dropping NETIF_F_TSO since no "
4914 "SG feature.\n", name
);
4915 features
&= ~NETIF_F_TSO
;
4918 if (features
& NETIF_F_UFO
) {
4919 if (!(features
& NETIF_F_GEN_CSUM
)) {
4921 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
4922 "since no NETIF_F_HW_CSUM feature.\n",
4924 features
&= ~NETIF_F_UFO
;
4927 if (!(features
& NETIF_F_SG
)) {
4929 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
4930 "since no NETIF_F_SG feature.\n", name
);
4931 features
&= ~NETIF_F_UFO
;
4937 EXPORT_SYMBOL(netdev_fix_features
);
4940 * netif_stacked_transfer_operstate - transfer operstate
4941 * @rootdev: the root or lower level device to transfer state from
4942 * @dev: the device to transfer operstate to
4944 * Transfer operational state from root to device. This is normally
4945 * called when a stacking relationship exists between the root
4946 * device and the device(a leaf device).
4948 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
4949 struct net_device
*dev
)
4951 if (rootdev
->operstate
== IF_OPER_DORMANT
)
4952 netif_dormant_on(dev
);
4954 netif_dormant_off(dev
);
4956 if (netif_carrier_ok(rootdev
)) {
4957 if (!netif_carrier_ok(dev
))
4958 netif_carrier_on(dev
);
4960 if (netif_carrier_ok(dev
))
4961 netif_carrier_off(dev
);
4964 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
4967 * register_netdevice - register a network device
4968 * @dev: device to register
4970 * Take a completed network device structure and add it to the kernel
4971 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4972 * chain. 0 is returned on success. A negative errno code is returned
4973 * on a failure to set up the device, or if the name is a duplicate.
4975 * Callers must hold the rtnl semaphore. You may want
4976 * register_netdev() instead of this.
4979 * The locking appears insufficient to guarantee two parallel registers
4980 * will not get the same name.
4983 int register_netdevice(struct net_device
*dev
)
4986 struct net
*net
= dev_net(dev
);
4988 BUG_ON(dev_boot_phase
);
4993 /* When net_device's are persistent, this will be fatal. */
4994 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
4997 spin_lock_init(&dev
->addr_list_lock
);
4998 netdev_set_addr_lockdep_class(dev
);
4999 netdev_init_queue_locks(dev
);
5004 if (!dev
->num_rx_queues
) {
5006 * Allocate a single RX queue if driver never called
5010 dev
->_rx
= kzalloc(sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5016 dev
->_rx
->first
= dev
->_rx
;
5017 atomic_set(&dev
->_rx
->count
, 1);
5018 dev
->num_rx_queues
= 1;
5021 /* Init, if this function is available */
5022 if (dev
->netdev_ops
->ndo_init
) {
5023 ret
= dev
->netdev_ops
->ndo_init(dev
);
5031 ret
= dev_get_valid_name(dev
, dev
->name
, 0);
5035 dev
->ifindex
= dev_new_index(net
);
5036 if (dev
->iflink
== -1)
5037 dev
->iflink
= dev
->ifindex
;
5039 /* Fix illegal checksum combinations */
5040 if ((dev
->features
& NETIF_F_HW_CSUM
) &&
5041 (dev
->features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5042 printk(KERN_NOTICE
"%s: mixed HW and IP checksum settings.\n",
5044 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5047 if ((dev
->features
& NETIF_F_NO_CSUM
) &&
5048 (dev
->features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5049 printk(KERN_NOTICE
"%s: mixed no checksumming and other settings.\n",
5051 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
5054 dev
->features
= netdev_fix_features(dev
->features
, dev
->name
);
5056 /* Enable software GSO if SG is supported. */
5057 if (dev
->features
& NETIF_F_SG
)
5058 dev
->features
|= NETIF_F_GSO
;
5060 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5061 ret
= notifier_to_errno(ret
);
5065 ret
= netdev_register_kobject(dev
);
5068 dev
->reg_state
= NETREG_REGISTERED
;
5071 * Default initial state at registry is that the
5072 * device is present.
5075 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5077 dev_init_scheduler(dev
);
5079 list_netdevice(dev
);
5081 /* Notify protocols, that a new device appeared. */
5082 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5083 ret
= notifier_to_errno(ret
);
5085 rollback_registered(dev
);
5086 dev
->reg_state
= NETREG_UNREGISTERED
;
5089 * Prevent userspace races by waiting until the network
5090 * device is fully setup before sending notifications.
5092 if (!dev
->rtnl_link_ops
||
5093 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5094 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5100 if (dev
->netdev_ops
->ndo_uninit
)
5101 dev
->netdev_ops
->ndo_uninit(dev
);
5104 EXPORT_SYMBOL(register_netdevice
);
5107 * init_dummy_netdev - init a dummy network device for NAPI
5108 * @dev: device to init
5110 * This takes a network device structure and initialize the minimum
5111 * amount of fields so it can be used to schedule NAPI polls without
5112 * registering a full blown interface. This is to be used by drivers
5113 * that need to tie several hardware interfaces to a single NAPI
5114 * poll scheduler due to HW limitations.
5116 int init_dummy_netdev(struct net_device
*dev
)
5118 /* Clear everything. Note we don't initialize spinlocks
5119 * are they aren't supposed to be taken by any of the
5120 * NAPI code and this dummy netdev is supposed to be
5121 * only ever used for NAPI polls
5123 memset(dev
, 0, sizeof(struct net_device
));
5125 /* make sure we BUG if trying to hit standard
5126 * register/unregister code path
5128 dev
->reg_state
= NETREG_DUMMY
;
5130 /* initialize the ref count */
5131 atomic_set(&dev
->refcnt
, 1);
5133 /* NAPI wants this */
5134 INIT_LIST_HEAD(&dev
->napi_list
);
5136 /* a dummy interface is started by default */
5137 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5138 set_bit(__LINK_STATE_START
, &dev
->state
);
5142 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5146 * register_netdev - register a network device
5147 * @dev: device to register
5149 * Take a completed network device structure and add it to the kernel
5150 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5151 * chain. 0 is returned on success. A negative errno code is returned
5152 * on a failure to set up the device, or if the name is a duplicate.
5154 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5155 * and expands the device name if you passed a format string to
5158 int register_netdev(struct net_device
*dev
)
5165 * If the name is a format string the caller wants us to do a
5168 if (strchr(dev
->name
, '%')) {
5169 err
= dev_alloc_name(dev
, dev
->name
);
5174 err
= register_netdevice(dev
);
5179 EXPORT_SYMBOL(register_netdev
);
5182 * netdev_wait_allrefs - wait until all references are gone.
5184 * This is called when unregistering network devices.
5186 * Any protocol or device that holds a reference should register
5187 * for netdevice notification, and cleanup and put back the
5188 * reference if they receive an UNREGISTER event.
5189 * We can get stuck here if buggy protocols don't correctly
5192 static void netdev_wait_allrefs(struct net_device
*dev
)
5194 unsigned long rebroadcast_time
, warning_time
;
5196 linkwatch_forget_dev(dev
);
5198 rebroadcast_time
= warning_time
= jiffies
;
5199 while (atomic_read(&dev
->refcnt
) != 0) {
5200 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5203 /* Rebroadcast unregister notification */
5204 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5205 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5206 * should have already handle it the first time */
5208 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5210 /* We must not have linkwatch events
5211 * pending on unregister. If this
5212 * happens, we simply run the queue
5213 * unscheduled, resulting in a noop
5216 linkwatch_run_queue();
5221 rebroadcast_time
= jiffies
;
5226 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5227 printk(KERN_EMERG
"unregister_netdevice: "
5228 "waiting for %s to become free. Usage "
5230 dev
->name
, atomic_read(&dev
->refcnt
));
5231 warning_time
= jiffies
;
5240 * register_netdevice(x1);
5241 * register_netdevice(x2);
5243 * unregister_netdevice(y1);
5244 * unregister_netdevice(y2);
5250 * We are invoked by rtnl_unlock().
5251 * This allows us to deal with problems:
5252 * 1) We can delete sysfs objects which invoke hotplug
5253 * without deadlocking with linkwatch via keventd.
5254 * 2) Since we run with the RTNL semaphore not held, we can sleep
5255 * safely in order to wait for the netdev refcnt to drop to zero.
5257 * We must not return until all unregister events added during
5258 * the interval the lock was held have been completed.
5260 void netdev_run_todo(void)
5262 struct list_head list
;
5264 /* Snapshot list, allow later requests */
5265 list_replace_init(&net_todo_list
, &list
);
5269 while (!list_empty(&list
)) {
5270 struct net_device
*dev
5271 = list_first_entry(&list
, struct net_device
, todo_list
);
5272 list_del(&dev
->todo_list
);
5274 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5275 printk(KERN_ERR
"network todo '%s' but state %d\n",
5276 dev
->name
, dev
->reg_state
);
5281 dev
->reg_state
= NETREG_UNREGISTERED
;
5283 on_each_cpu(flush_backlog
, dev
, 1);
5285 netdev_wait_allrefs(dev
);
5288 BUG_ON(atomic_read(&dev
->refcnt
));
5289 WARN_ON(dev
->ip_ptr
);
5290 WARN_ON(dev
->ip6_ptr
);
5291 WARN_ON(dev
->dn_ptr
);
5293 if (dev
->destructor
)
5294 dev
->destructor(dev
);
5296 /* Free network device */
5297 kobject_put(&dev
->dev
.kobj
);
5302 * dev_txq_stats_fold - fold tx_queues stats
5303 * @dev: device to get statistics from
5304 * @stats: struct rtnl_link_stats64 to hold results
5306 void dev_txq_stats_fold(const struct net_device
*dev
,
5307 struct rtnl_link_stats64
*stats
)
5309 u64 tx_bytes
= 0, tx_packets
= 0, tx_dropped
= 0;
5311 struct netdev_queue
*txq
;
5313 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
5314 txq
= netdev_get_tx_queue(dev
, i
);
5315 spin_lock_bh(&txq
->_xmit_lock
);
5316 tx_bytes
+= txq
->tx_bytes
;
5317 tx_packets
+= txq
->tx_packets
;
5318 tx_dropped
+= txq
->tx_dropped
;
5319 spin_unlock_bh(&txq
->_xmit_lock
);
5321 if (tx_bytes
|| tx_packets
|| tx_dropped
) {
5322 stats
->tx_bytes
= tx_bytes
;
5323 stats
->tx_packets
= tx_packets
;
5324 stats
->tx_dropped
= tx_dropped
;
5327 EXPORT_SYMBOL(dev_txq_stats_fold
);
5329 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5330 * fields in the same order, with only the type differing.
5332 static void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
5333 const struct net_device_stats
*netdev_stats
)
5335 #if BITS_PER_LONG == 64
5336 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
5337 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
5339 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
5340 const unsigned long *src
= (const unsigned long *)netdev_stats
;
5341 u64
*dst
= (u64
*)stats64
;
5343 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
5344 sizeof(*stats64
) / sizeof(u64
));
5345 for (i
= 0; i
< n
; i
++)
5351 * dev_get_stats - get network device statistics
5352 * @dev: device to get statistics from
5353 * @storage: place to store stats
5355 * Get network statistics from device. Return @storage.
5356 * The device driver may provide its own method by setting
5357 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5358 * otherwise the internal statistics structure is used.
5360 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
5361 struct rtnl_link_stats64
*storage
)
5363 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5365 if (ops
->ndo_get_stats64
) {
5366 memset(storage
, 0, sizeof(*storage
));
5367 return ops
->ndo_get_stats64(dev
, storage
);
5369 if (ops
->ndo_get_stats
) {
5370 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
5373 netdev_stats_to_stats64(storage
, &dev
->stats
);
5374 dev_txq_stats_fold(dev
, storage
);
5377 EXPORT_SYMBOL(dev_get_stats
);
5379 static void netdev_init_one_queue(struct net_device
*dev
,
5380 struct netdev_queue
*queue
,
5386 static void netdev_init_queues(struct net_device
*dev
)
5388 netdev_init_one_queue(dev
, &dev
->rx_queue
, NULL
);
5389 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5390 spin_lock_init(&dev
->tx_global_lock
);
5394 * alloc_netdev_mq - allocate network device
5395 * @sizeof_priv: size of private data to allocate space for
5396 * @name: device name format string
5397 * @setup: callback to initialize device
5398 * @queue_count: the number of subqueues to allocate
5400 * Allocates a struct net_device with private data area for driver use
5401 * and performs basic initialization. Also allocates subquue structs
5402 * for each queue on the device at the end of the netdevice.
5404 struct net_device
*alloc_netdev_mq(int sizeof_priv
, const char *name
,
5405 void (*setup
)(struct net_device
*), unsigned int queue_count
)
5407 struct netdev_queue
*tx
;
5408 struct net_device
*dev
;
5410 struct net_device
*p
;
5412 struct netdev_rx_queue
*rx
;
5416 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5418 alloc_size
= sizeof(struct net_device
);
5420 /* ensure 32-byte alignment of private area */
5421 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5422 alloc_size
+= sizeof_priv
;
5424 /* ensure 32-byte alignment of whole construct */
5425 alloc_size
+= NETDEV_ALIGN
- 1;
5427 p
= kzalloc(alloc_size
, GFP_KERNEL
);
5429 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
5433 tx
= kcalloc(queue_count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5435 printk(KERN_ERR
"alloc_netdev: Unable to allocate "
5441 rx
= kcalloc(queue_count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5443 printk(KERN_ERR
"alloc_netdev: Unable to allocate "
5448 atomic_set(&rx
->count
, queue_count
);
5451 * Set a pointer to first element in the array which holds the
5454 for (i
= 0; i
< queue_count
; i
++)
5458 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
5459 dev
->padded
= (char *)dev
- (char *)p
;
5461 if (dev_addr_init(dev
))
5467 dev_net_set(dev
, &init_net
);
5470 dev
->num_tx_queues
= queue_count
;
5471 dev
->real_num_tx_queues
= queue_count
;
5475 dev
->num_rx_queues
= queue_count
;
5478 dev
->gso_max_size
= GSO_MAX_SIZE
;
5480 netdev_init_queues(dev
);
5482 INIT_LIST_HEAD(&dev
->ethtool_ntuple_list
.list
);
5483 dev
->ethtool_ntuple_list
.count
= 0;
5484 INIT_LIST_HEAD(&dev
->napi_list
);
5485 INIT_LIST_HEAD(&dev
->unreg_list
);
5486 INIT_LIST_HEAD(&dev
->link_watch_list
);
5487 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
5489 strcpy(dev
->name
, name
);
5502 EXPORT_SYMBOL(alloc_netdev_mq
);
5505 * free_netdev - free network device
5508 * This function does the last stage of destroying an allocated device
5509 * interface. The reference to the device object is released.
5510 * If this is the last reference then it will be freed.
5512 void free_netdev(struct net_device
*dev
)
5514 struct napi_struct
*p
, *n
;
5516 release_net(dev_net(dev
));
5520 /* Flush device addresses */
5521 dev_addr_flush(dev
);
5523 /* Clear ethtool n-tuple list */
5524 ethtool_ntuple_flush(dev
);
5526 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
5529 /* Compatibility with error handling in drivers */
5530 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5531 kfree((char *)dev
- dev
->padded
);
5535 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
5536 dev
->reg_state
= NETREG_RELEASED
;
5538 /* will free via device release */
5539 put_device(&dev
->dev
);
5541 EXPORT_SYMBOL(free_netdev
);
5544 * synchronize_net - Synchronize with packet receive processing
5546 * Wait for packets currently being received to be done.
5547 * Does not block later packets from starting.
5549 void synchronize_net(void)
5554 EXPORT_SYMBOL(synchronize_net
);
5557 * unregister_netdevice_queue - remove device from the kernel
5561 * This function shuts down a device interface and removes it
5562 * from the kernel tables.
5563 * If head not NULL, device is queued to be unregistered later.
5565 * Callers must hold the rtnl semaphore. You may want
5566 * unregister_netdev() instead of this.
5569 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
5574 list_move_tail(&dev
->unreg_list
, head
);
5576 rollback_registered(dev
);
5577 /* Finish processing unregister after unlock */
5581 EXPORT_SYMBOL(unregister_netdevice_queue
);
5584 * unregister_netdevice_many - unregister many devices
5585 * @head: list of devices
5587 void unregister_netdevice_many(struct list_head
*head
)
5589 struct net_device
*dev
;
5591 if (!list_empty(head
)) {
5592 rollback_registered_many(head
);
5593 list_for_each_entry(dev
, head
, unreg_list
)
5597 EXPORT_SYMBOL(unregister_netdevice_many
);
5600 * unregister_netdev - remove device from the kernel
5603 * This function shuts down a device interface and removes it
5604 * from the kernel tables.
5606 * This is just a wrapper for unregister_netdevice that takes
5607 * the rtnl semaphore. In general you want to use this and not
5608 * unregister_netdevice.
5610 void unregister_netdev(struct net_device
*dev
)
5613 unregister_netdevice(dev
);
5616 EXPORT_SYMBOL(unregister_netdev
);
5619 * dev_change_net_namespace - move device to different nethost namespace
5621 * @net: network namespace
5622 * @pat: If not NULL name pattern to try if the current device name
5623 * is already taken in the destination network namespace.
5625 * This function shuts down a device interface and moves it
5626 * to a new network namespace. On success 0 is returned, on
5627 * a failure a netagive errno code is returned.
5629 * Callers must hold the rtnl semaphore.
5632 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
5638 /* Don't allow namespace local devices to be moved. */
5640 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
5643 /* Ensure the device has been registrered */
5645 if (dev
->reg_state
!= NETREG_REGISTERED
)
5648 /* Get out if there is nothing todo */
5650 if (net_eq(dev_net(dev
), net
))
5653 /* Pick the destination device name, and ensure
5654 * we can use it in the destination network namespace.
5657 if (__dev_get_by_name(net
, dev
->name
)) {
5658 /* We get here if we can't use the current device name */
5661 if (dev_get_valid_name(dev
, pat
, 1))
5666 * And now a mini version of register_netdevice unregister_netdevice.
5669 /* If device is running close it first. */
5672 /* And unlink it from device chain */
5674 unlist_netdevice(dev
);
5678 /* Shutdown queueing discipline. */
5681 /* Notify protocols, that we are about to destroy
5682 this device. They should clean all the things.
5684 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5685 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5688 * Flush the unicast and multicast chains
5693 /* Actually switch the network namespace */
5694 dev_net_set(dev
, net
);
5696 /* If there is an ifindex conflict assign a new one */
5697 if (__dev_get_by_index(net
, dev
->ifindex
)) {
5698 int iflink
= (dev
->iflink
== dev
->ifindex
);
5699 dev
->ifindex
= dev_new_index(net
);
5701 dev
->iflink
= dev
->ifindex
;
5704 /* Fixup kobjects */
5705 err
= device_rename(&dev
->dev
, dev
->name
);
5708 /* Add the device back in the hashes */
5709 list_netdevice(dev
);
5711 /* Notify protocols, that a new device appeared. */
5712 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5715 * Prevent userspace races by waiting until the network
5716 * device is fully setup before sending notifications.
5718 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5725 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
5727 static int dev_cpu_callback(struct notifier_block
*nfb
,
5728 unsigned long action
,
5731 struct sk_buff
**list_skb
;
5732 struct sk_buff
*skb
;
5733 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
5734 struct softnet_data
*sd
, *oldsd
;
5736 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
5739 local_irq_disable();
5740 cpu
= smp_processor_id();
5741 sd
= &per_cpu(softnet_data
, cpu
);
5742 oldsd
= &per_cpu(softnet_data
, oldcpu
);
5744 /* Find end of our completion_queue. */
5745 list_skb
= &sd
->completion_queue
;
5747 list_skb
= &(*list_skb
)->next
;
5748 /* Append completion queue from offline CPU. */
5749 *list_skb
= oldsd
->completion_queue
;
5750 oldsd
->completion_queue
= NULL
;
5752 /* Append output queue from offline CPU. */
5753 if (oldsd
->output_queue
) {
5754 *sd
->output_queue_tailp
= oldsd
->output_queue
;
5755 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
5756 oldsd
->output_queue
= NULL
;
5757 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
5760 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
5763 /* Process offline CPU's input_pkt_queue */
5764 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
5766 input_queue_head_incr(oldsd
);
5768 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
5770 input_queue_head_incr(oldsd
);
5778 * netdev_increment_features - increment feature set by one
5779 * @all: current feature set
5780 * @one: new feature set
5781 * @mask: mask feature set
5783 * Computes a new feature set after adding a device with feature set
5784 * @one to the master device with current feature set @all. Will not
5785 * enable anything that is off in @mask. Returns the new feature set.
5787 unsigned long netdev_increment_features(unsigned long all
, unsigned long one
,
5790 /* If device needs checksumming, downgrade to it. */
5791 if (all
& NETIF_F_NO_CSUM
&& !(one
& NETIF_F_NO_CSUM
))
5792 all
^= NETIF_F_NO_CSUM
| (one
& NETIF_F_ALL_CSUM
);
5793 else if (mask
& NETIF_F_ALL_CSUM
) {
5794 /* If one device supports v4/v6 checksumming, set for all. */
5795 if (one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
) &&
5796 !(all
& NETIF_F_GEN_CSUM
)) {
5797 all
&= ~NETIF_F_ALL_CSUM
;
5798 all
|= one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
);
5801 /* If one device supports hw checksumming, set for all. */
5802 if (one
& NETIF_F_GEN_CSUM
&& !(all
& NETIF_F_GEN_CSUM
)) {
5803 all
&= ~NETIF_F_ALL_CSUM
;
5804 all
|= NETIF_F_HW_CSUM
;
5808 one
|= NETIF_F_ALL_CSUM
;
5810 one
|= all
& NETIF_F_ONE_FOR_ALL
;
5811 all
&= one
| NETIF_F_LLTX
| NETIF_F_GSO
| NETIF_F_UFO
;
5812 all
|= one
& mask
& NETIF_F_ONE_FOR_ALL
;
5816 EXPORT_SYMBOL(netdev_increment_features
);
5818 static struct hlist_head
*netdev_create_hash(void)
5821 struct hlist_head
*hash
;
5823 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
5825 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
5826 INIT_HLIST_HEAD(&hash
[i
]);
5831 /* Initialize per network namespace state */
5832 static int __net_init
netdev_init(struct net
*net
)
5834 INIT_LIST_HEAD(&net
->dev_base_head
);
5836 net
->dev_name_head
= netdev_create_hash();
5837 if (net
->dev_name_head
== NULL
)
5840 net
->dev_index_head
= netdev_create_hash();
5841 if (net
->dev_index_head
== NULL
)
5847 kfree(net
->dev_name_head
);
5853 * netdev_drivername - network driver for the device
5854 * @dev: network device
5855 * @buffer: buffer for resulting name
5856 * @len: size of buffer
5858 * Determine network driver for device.
5860 char *netdev_drivername(const struct net_device
*dev
, char *buffer
, int len
)
5862 const struct device_driver
*driver
;
5863 const struct device
*parent
;
5865 if (len
<= 0 || !buffer
)
5869 parent
= dev
->dev
.parent
;
5874 driver
= parent
->driver
;
5875 if (driver
&& driver
->name
)
5876 strlcpy(buffer
, driver
->name
, len
);
5880 static int __netdev_printk(const char *level
, const struct net_device
*dev
,
5881 struct va_format
*vaf
)
5885 if (dev
&& dev
->dev
.parent
)
5886 r
= dev_printk(level
, dev
->dev
.parent
, "%s: %pV",
5887 netdev_name(dev
), vaf
);
5889 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
5891 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
5896 int netdev_printk(const char *level
, const struct net_device
*dev
,
5897 const char *format
, ...)
5899 struct va_format vaf
;
5903 va_start(args
, format
);
5908 r
= __netdev_printk(level
, dev
, &vaf
);
5913 EXPORT_SYMBOL(netdev_printk
);
5915 #define define_netdev_printk_level(func, level) \
5916 int func(const struct net_device *dev, const char *fmt, ...) \
5919 struct va_format vaf; \
5922 va_start(args, fmt); \
5927 r = __netdev_printk(level, dev, &vaf); \
5932 EXPORT_SYMBOL(func);
5934 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
5935 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
5936 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
5937 define_netdev_printk_level(netdev_err
, KERN_ERR
);
5938 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
5939 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
5940 define_netdev_printk_level(netdev_info
, KERN_INFO
);
5942 static void __net_exit
netdev_exit(struct net
*net
)
5944 kfree(net
->dev_name_head
);
5945 kfree(net
->dev_index_head
);
5948 static struct pernet_operations __net_initdata netdev_net_ops
= {
5949 .init
= netdev_init
,
5950 .exit
= netdev_exit
,
5953 static void __net_exit
default_device_exit(struct net
*net
)
5955 struct net_device
*dev
, *aux
;
5957 * Push all migratable network devices back to the
5958 * initial network namespace
5961 for_each_netdev_safe(net
, dev
, aux
) {
5963 char fb_name
[IFNAMSIZ
];
5965 /* Ignore unmoveable devices (i.e. loopback) */
5966 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
5969 /* Leave virtual devices for the generic cleanup */
5970 if (dev
->rtnl_link_ops
)
5973 /* Push remaing network devices to init_net */
5974 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
5975 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
5977 printk(KERN_EMERG
"%s: failed to move %s to init_net: %d\n",
5978 __func__
, dev
->name
, err
);
5985 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
5987 /* At exit all network devices most be removed from a network
5988 * namespace. Do this in the reverse order of registeration.
5989 * Do this across as many network namespaces as possible to
5990 * improve batching efficiency.
5992 struct net_device
*dev
;
5994 LIST_HEAD(dev_kill_list
);
5997 list_for_each_entry(net
, net_list
, exit_list
) {
5998 for_each_netdev_reverse(net
, dev
) {
5999 if (dev
->rtnl_link_ops
)
6000 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6002 unregister_netdevice_queue(dev
, &dev_kill_list
);
6005 unregister_netdevice_many(&dev_kill_list
);
6009 static struct pernet_operations __net_initdata default_device_ops
= {
6010 .exit
= default_device_exit
,
6011 .exit_batch
= default_device_exit_batch
,
6015 * Initialize the DEV module. At boot time this walks the device list and
6016 * unhooks any devices that fail to initialise (normally hardware not
6017 * present) and leaves us with a valid list of present and active devices.
6022 * This is called single threaded during boot, so no need
6023 * to take the rtnl semaphore.
6025 static int __init
net_dev_init(void)
6027 int i
, rc
= -ENOMEM
;
6029 BUG_ON(!dev_boot_phase
);
6031 if (dev_proc_init())
6034 if (netdev_kobject_init())
6037 INIT_LIST_HEAD(&ptype_all
);
6038 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6039 INIT_LIST_HEAD(&ptype_base
[i
]);
6041 if (register_pernet_subsys(&netdev_net_ops
))
6045 * Initialise the packet receive queues.
6048 for_each_possible_cpu(i
) {
6049 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6051 memset(sd
, 0, sizeof(*sd
));
6052 skb_queue_head_init(&sd
->input_pkt_queue
);
6053 skb_queue_head_init(&sd
->process_queue
);
6054 sd
->completion_queue
= NULL
;
6055 INIT_LIST_HEAD(&sd
->poll_list
);
6056 sd
->output_queue
= NULL
;
6057 sd
->output_queue_tailp
= &sd
->output_queue
;
6059 sd
->csd
.func
= rps_trigger_softirq
;
6065 sd
->backlog
.poll
= process_backlog
;
6066 sd
->backlog
.weight
= weight_p
;
6067 sd
->backlog
.gro_list
= NULL
;
6068 sd
->backlog
.gro_count
= 0;
6073 /* The loopback device is special if any other network devices
6074 * is present in a network namespace the loopback device must
6075 * be present. Since we now dynamically allocate and free the
6076 * loopback device ensure this invariant is maintained by
6077 * keeping the loopback device as the first device on the
6078 * list of network devices. Ensuring the loopback devices
6079 * is the first device that appears and the last network device
6082 if (register_pernet_device(&loopback_net_ops
))
6085 if (register_pernet_device(&default_device_ops
))
6088 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6089 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6091 hotcpu_notifier(dev_cpu_callback
, 0);
6099 subsys_initcall(net_dev_init
);
6101 static int __init
initialize_hashrnd(void)
6103 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
6107 late_initcall_sync(initialize_hashrnd
);